A Psycholinguistic Study
of Temporary Ambiguity Resolution
in Sentence Processing
Canice Grant
B.A. (Mod.) CSLL
Final Year Project
May 2005
Supervisor: Dr. Martin Emms
Declaration
I hereby declare that this thesis is entirely my own work and that it has not been submitted
as an exercise for a degree at any other university
____________________________________ April 15, 2005
Canice Grant
2
Acknowledgements
I would like to thank my supervisor Dr. Martin Emms who was always helpful and patient.
His advice was welcomed and appreciated.
I would also like to thank my parents, Teresa and Michael Grant, for giving me the
opportunity and support needed to achieve personal goals.
I would also like to thank Yvette Graham, the experiment manager, who was extremely
helpful in setting up the on-line experiment.
Finally, I would like to thank all my friends who listened to me throughout and offered their
valued opinions, especially Lauren Breatnach.
Contents
1.
Introduction
1.1 Introduction………………………………………………………………………
1.2 Background………………………………………………………………………
10
10
3
1.3 Motivations and Aims of the Project…………………………………………….
1.4 Psycholinguistics…………………………………………………………………
1.5 Structure of the Report…………………………………………………………...
2.
3.
4.
2.1
2.2
2.3
2.4
11
12
13
Temporary Syntactic Ambiguity
Introduction………………………………………………………………………
What is Syntax?………………………………………………………………….
The Role of Syntax in a Language……………………………………………….
Ambiguity………………………………………………………………………..
2.4.1
Syntactic Ambiguity……………………………………………………
2.4.2
Temporary Ambiguity………………………………………………….
2.5 Temporary Syntactic Ambiguity in Language…………………………………..
17
17
19
20
21
23
25
Sentence Processing
3.1 Introduction……………………………………………………………………….
27
3.2 The Role of Syntax in Sentence Processing………………………………………
3.3 The Syntactic Challenges in Processing Written Sentences………………………
3.3.1
Word Order………………………………………………………………
3.3.2
Minimum-distance Principle…………………………………………….
3.3.3
Coined-clause Analysis………………………………………………….
3.3.4
Passive Voice……………………………………………………………
3.3.5
Negation…………………………………………………………………
3.3.6
Embedding………………………………………………………………
3.4 The Role of Working Memory in Sentence Comprehension…………………….
3.5 The Processing Theories………………………………………………………….
3.5.1
The Serial Model………………………………………………………..
3.5.2
The Parallel Model………………………………………………………
27
28
28
29
29
30
31
31
32
34
34
36
Methodology
4.1 Introduction……………………………………………………………………….
4.2 Experiment Parameters……………………………………………………………
4.2.1
Selection of Subjects……………………………………………………
4.2.2
The Instructions…………………………………………………………
4.3 Temporary Syntactic Ambiguity Categories………………………………………
4.3.1 Complement Clause Vs Relative Clause…………………………………
4.3.2 Object Noun Phrase Vs Main Clause Subject……………………………
4.3.3 Direct Object Vs Embedded Subject…………………………………….
4.3.4 Prepositional Phrase Attachement……………………………………….
4.3.4 Main Clause Vs Complement Clause……………………………………
4.4 ICECUP Version 3.0……………………………………………………………...
4.4.1 Disadvantages…………………………………………………………….
4.5 Design of the Experiment…………………………………………………………
4.5.1 Five Versions…………………………………………………………….
4.5.2 The Instructions………………………………………………………….
4.5.3 The Sentences……………………………………………………………
4.5.4 The Participants………………………………………………………….
5.
40
40
41
42
42
43
45
47
48
49
50
53
55
56
57
58
59
Experimentation
5.1 Introduction…………………………………………………………………….
5.2 Web Tool……………………………………………………………………….
5.3 Using the Tool to Define the Experiment………………………………………
5.3.1
The Ambi-pair Design………………………………………………..
5.3.2
The Context Paragraph Design……………………………………….
5.3.3
The Form Design……………………………………………………..
61
61
63
65
66
68
4
5.4 Participation…………………………………………………………………….
5.5 Viewing the Results…………………………………………………………….
5.5.1
The Temporal Analysis………………………………………………..
5.6 Problems Encountered…………………………………………………………..
5.7 Possible Improvements………………………………………………………….
6.
Results
6.1 Introduction………………………………………………………………………
6.2 Structure of Results………………………………………………………………
6.3 Complement Clause Vs Relative Clause…………………………………………
6.4 Object Noun Phrase Vs Main Clause Subject…………………………………….
6.5 Direct Object Vs Embedded Subject……………………………………………..
6.6 Prepositional Phrase Attachment…………………………………………………
6.7 Main Clause Vs Complement Clause……………………………………………
6.8 Context Paragraphs………………………………………………………………
6.9 Conclusion……………………………………………………………………….
7.
7.1
7.2
7.3
7.4
7.5
7.6
7.7
8.
69
70
71
73
74
8.1
8.2
8.3
8.4
8.5
77
79
81
83
85
87
89
91
94
Analysis of the Results
Introduction………………………………………………………………………
Temporary Ambiguity Categories……………………………………………….
Male Vs Female………………………………………………………………….
Linguists Vs Non-linguists……………………………………………………….
Bilinguals Vs Native Speakers…………………………………………………..
Processing Theories Revisited……………………………………………………
Discussion………………………………………………………………………..
7.7.1
The Ambi-pairs………………………………………………………..
7.7.2
The Context Paragraphs…………………………………………………
96
96
98
100
101
103
108
108
109
Conclusion
Introduction………………………………………………………………………
Achievements and Knowledge Gained………………………………………….
Difficulties Encountered…………………………………………………………
Suggestions for Further Study…………………………………………………..
Summary…………………………………………………………………………
112
112
113
113
113
Bibliography…………………………………………..
114
Appendix…………………………………………………
117
Experiment Instructions….………………………………………………………..
118
Experiment Sentences…..………………………………………………………….
Temporary Ambiguity Results……………………………………………………..
118
124
Abstract
5
This paper describes a psycholinguistic study that has been undertaken to
investigate the human resolution of temporary ambiguity. It describes a web-based
experimental tool, which has been designed to evaluate the ways in which humans treat
temporary ambiguity, the different types of ambiguity and the problems related to the
processing of such ambiguity. The treatment of these complexities in language is compared
to those associated with bilingualism and considers whether previous linguistic instruction
plays an important role in this process. The actual make-up of the human language
processor is also investigated and looks at the ways in which the study of temporary
ambiguity may inform us, in terms of the model that most resembles the human faculty.
List of figures
1.1.1
Syntax tree
2.3.1.1
Ambiguous syntax tree
2.3.1.2
Ambiguous syntax tree
2.3.2.1
Temporary syntactic ambiguity tree
3.5.1.1
Serial model diagram
3.5.1.2
Serial model language processor diagram
3.5.2.1
Parallel model diagram
3.5.2.2 Parallel model language processor diagram
6
4.2.2.1
Object noun phrase Vs Main clause subject syntax tree
4.2.2.2
Object noun phrase Vs Main clause subject syntax tree
4.2.4.1
Prepositional phrase syntax tree
4.4.1
ICECUP syntax tree
4.4.2
ICECUP results
4.4.3
ICECUP word search
4.4.1.1
ICECUP irregularities
4.5.1.1
Experiment layout
5.3.1
Web tool homepage
5.3.1.1
Ambi-pair web tool design
5.3.1.2
Web tool slide page
5.3.2.1
Web tool question page
5.3.3.1
Web tool form page
5.4.1
Slip of paper
5.5.1
Web tool result directory page
5.5.1.1
A typical “.csv” file
5.5.1.2
An Excel platform for a “.csv” file
6.1.1
Frequency of participation
6.1.2
Participation age
6.1.3
Participant sex
6.1.4
Linguistic orientation
6.1.5
Bilingual participation
6.3.1, 6.4.1, 6.5.1, 6.6.1, 6.7.1 Ambi-pair control question results per subject
6.3.2, 6.4.2, 6.5.2, 6.6.2, 6.7.2
Ambi-pair frequency of answers
6.8.1
Context Vs Isolated control question results
6.8.2
Context Vs Isolated response times
7.2.1
Category response times
7.2.2
Category control answers
7.3.1, 7.4.1, 7.5.1
Analysis response times
7.3.2, 7.4.2, 7.5.2
Analysis control answers
7.6.1
Context Vs Isolated response times
7.6.2
Analysis response times
7.6.3
Category context Vs Isolated response times
7.6.4
Context Vs Isolated control answers
7.6.5
Analysis Context Vs Isolated control answers
7.6.6
Category context Vs Isolated control questions
7
“The basic task of readers is similar to the task of a prospector. Just as the prospector
picks away at the surface to discover the gold hidden underneath, readers dig away at the
surface structure, searching for and demanding meaning.”
‘Helping Children Learn to Read’, Searfoss and Readence
Chapter 1
Introduction
8
1.1
Introduction
This chapter is an introduction to my final year project, which is concerned with
the resolution of temporary ambiguity. Using a web-based tool to conduct an experiment, I
hope to establish certain facts about the nature of the different types of temporary
ambiguity and the difficulties faced in resolving such phenomena during the processing of
sentences. In addition, I hope to be able to infer some of the properties of the human
processing faculty. The background to the project is explained in section 1.2 with the
motivation behind the paper and the overall aims being covered in section 1.3. A
background in psycholinguistics is provided in section 1.4 which is followed by an
overview and structure to the remainder of the project in section 1.5.
9
1.2
Background
Ambiguity can be found under many guises and in many different contexts. It has
often been used by scientists in order to verify or contradict certain hypotheses as to how
the human language faculty actually functions. In this paper and by means of
experimentation, I hope to further understand the underlying concepts behind temporary
ambiguity. It must be noted that temporary ambiguity is often studied under the concept
known as “Garden Paths”. See Crain&Steedman (1985) in which a substantial body of
work has been accomplished on this particular topic. This study also considers the different
types of temporary ambiguity, how they are related to each other and the difficulties faced
whilst processing sentences containing such ambiguity. It also attempts to define and
categorise those people who find the resolution of such sentences more difficult than others.
Experiments are used by linguists to prove or disprove certain hypotheses. Native
speakers often play their part in these discoveries. Whorf (1957) stated that:
“scientific linguists have long understood that ability to speak a language fluently does
not necessarily confer a linguistic knowledge of it.”
The intuitive construction of sentences by native speakers is one of the factors,
which promotes effective experiments to advance the understanding of human linguistic
processes. The subject of the experiment is generally unaware of that which is being tested
and is therefore unable to influence the results. The intuition of native speakers can also be
used in contrast to those of bilinguals. The language that bilinguals possess is clearly
different and particular differences can be uncovered by the effective use of experimental
tools.
Following research on the topics of “Garden Paths” and sentence processing, an
interesting experiment was conducted to establish how the two concepts intertwine and to
how they can be used to clarify hypotheses to which they relate. The difficulties that were
experienced in conducting the experiment were mitigated, to some extent, with the aid of
computers. Several tools were used on the computer in order to gather material and run the
final experiment.
A corpus based software tool was employed to gather the relevant material for the
experiment. It is described in section 4.4 but at this stage, I will simply describe it as a
useful tool that has been developed over the past fifteen years that can be applied to a wide
range of applications. A web based tool was also employed.
The generic web-based experimental tool used to conduct the experiment was one
previously developed for a final year project by several past graduates of Computer
Science, Linguistics and a Language. It was more recently improved upon by Fionnuala
Hourihane (2002) and is currently being refined by Yvette Graham, a masters student in
computational linguistics. This programme allowed me to set up an experiment in which I
10
could test peoples’ reaction to temporary ambiguity and determine how quickly they could
resolve different sentence constructions.
The third piece of software was used in processing to the results. The temporal analysis of
Chapter Six used a Microsoft Excel spreadsheet to create graphs. It is an extremely diverse
programme that can accommodate a variety of uses.
1.3
Motivations and Aims of the Project
Four years of studying Computer Science, Linguistics and French has, I believe,
taught me a great deal about language as a means of communication. I have studied the
mechanisms involved in learning and understanding language and have found that the
methods by which we communicate are very complex. Human language is incredibly
intricate and it is this level of intricacy that so fascinates me. Ambiguity in language is also
a common phenomenon. The ways in which people communicate with each other, given
their varying levels of ability, is of great interest to me. For instance, why are some people
able to resolve structures faster than others or, why does one particular type of ambiguity
produce an overall increase in difficulty when compared to another. These are some of the
questions, which I have sought to answer and are the motivating factors in my undertaking
this project. Whilst studying these aspects of ambiguity, it adds to my overall understanding
of how the language faculty functions.
Language can also be described as being extremely diverse among our species and
it is the learning of this diversity that I find so enjoyable. Polyglotism is, therefore, an area
of great interest to me. The differences that are evident between the language of native
speakers and learners of that language is the motivation for studying the way in which
bilinguals approach such pitfalls as temporary ambiguity.
The quirks and idioms of the different languages of the world has fascinated me since my
study of language began.
Psycholinguistics is, I believe, one of the most important aspects of linguistics and
this particular area of research has captivated me throughout my studies. To understand the
faculty that allows us to communicate through language is another motivation behind this
project. The study of how the brain is used to comprehend is an essential part of language
communication. It is widely believed, as discussed in Chapter Four, that there are two
distinct loci of computation in sentence processing (see Michael Niv 1993). There is a great
deal of controversy surrounding which of these two models is responsible for resolving
ambiguity. It is the aim of this project to acknowledge which of these is the more plausible
model for human processing. The aim of this paper is to encompass a greater overall
understanding of ambiguity and how this concept is treated by us during processing. It
should, therefore, be possible to draw certain conclusions as to what is involved in the
processing of sentences.
An answer to the question of whether previous formal instruction in linguistics can
alter the way in which we process information is also an aim of this project. More
specifically, can linguists resolve complexities in language more efficiently than those who
have not been formally taught the underlying structures.
11
1.4 Psycholinguistics
Psycholinguistics is a very important area of study. As its name suggests,
psycholinguistics is a discipline that links psychology and linguistics. In other words, it is
basically the study of how our minds process language. More specifically, it is:
The study of the mental faculties involved in the perception, production, and acquisition of
language1
As you might expect, the psycholinguistic discipline is extremely large and can vary from
areas concerned with speech therapy to those involving brain surgery. Caron (1992)
suggests a definition of psycholinguistics as the interdisciplinary experimental study of the
psychological processes through which a human implements the system of a natural
language, the relationship of language to human thought.
The goal of psycholinguistics is to find the most comprehensive and unified theory of
language behaviour. It is also concerned with the discovery of structures and processes
which underlie the human ability to produce and perceive natural language. Production can
therefore be studied in contrast to acquisition and it is under these two headings that many
psycholinguistic studies have taken place.
Differences and similarities discovered between the two, can alleviate many communicative
problems from which members of our species suffer. A sophisticated means of
communication is the greatest advantage separating us from every other animal on the
planet. Language is so complex and diverse that for every psycholinguistic study providing
the answer to one question, another two can be postulated as a result. Psycholinguistics is
striving to reduce the unknowns each and every day.
Psycholinguistics can be concentrated on either the theoretical side or the practical side.
The theoretical side is concerned with explaining the nature of language and how it is
acquired. The practical or applied perspective side of the discipline, attempts to apply
certain linguistic and psychological knowledge to problems such as reading, polyglotism,
and teaching. This involves producing hypotheses about language and then experimentally
testing these predictions, thus maintaining its claim to scientific status.
This paper is more concerned with the practical side of psycholinguistics. An experiment
was conducted to further understand the processes involved in the perception of language.
How these processes may differ across age and proficiency is also considered.
It must be mentioned that the study of psycholinguistics interacts with all of the social
sciences and the humanities. This is because it is such an intricate part of what makes us
human. As Kess (1992) suggests, psycholinguistics has far reaching connections with
disciplines such as philosophy, anthropology, sociology, communications engineering and
more recently, artificial intelligence. The importance of psycholinguistics is evident.
1
Definition from Merriam Webster on-line dictionary
12
1.5 Structure of the Report
Chapter Two: Temporary Syntactic Ambiguity
This chapter is primarily concerned with research into the current experiment. It more
specifically defines temporary syntactic ambiguity, a key notion of the paper. It begins by
generally defining what syntax in language is, and the role it plays in Section 2.2 and 2.3.
Ambiguity is then treated in section 2.4 and this, is in turn, related to temporary syntactic
ambiguity in the subsequent subsections.
Chapter Three: Sentence Processing
This chapter is concerned with the research behind the project and explains the principles
involved in the experiment. In Section 3.1, the role of syntax is considered with respect to
processing. This is followed in Section 3.3 by the background to general syntactic
difficulties that may present themselves during processing. Section 3.4 examines other
important aspects of sentence processing i.e. the working memory. Section 3.5 is a very
important section to the paper and describes the two hypothesised models of sentence
processing, serially or in parallel. These two models will be discussed again in Chapter
Seven during the analysis of the results. It represents part of the motivation behind the
experiment.
Chapter Four: Methodology
Chapter Four examines the methodology of the psycholinguistic experiment that this paper
describes. In Section 4.2, the categories of ambiguity that will be used in the experiment are
described. This is followed by a discussion in Section 4.3, outlining the general premises
that psycholinguistic experiments should follow. Schutze (1996) offers recommendation
parameters that can be adopted. In Section 4.4, the means of gathering material for the
experiment is discussed. A software programme called ICECUP is used and its
disadvantages are also debated. The design of the experiment is discussed in the final
section of this chapter and the individual aspects that coordinate to produce the final
experiment are discussed here.
Chapter Five: Experimentation
13
This chapter deals with the way in which the experiment is implemented. The web based
experimental tool is described in Section 5.2 and an overview of its workings is also
discussed. Section 5.3 relates to the design of the experiment and Chapter Four explores
the practical use of the web tool. The subsequent section explains how participation in the
experiment is achieved. In Section 5.5, the extraction of the results from the tool is
described. This subsection is concerned with the necessary changes that were made to the
tool in order to extract the relevant results. The final two sections of the chapter discuss the
problems encountered with the tool and possible future improvements to it.
Chapter Six: Results
This chapter considers the results that were extracted from the experiment. The
introduction treats the overall participation in the project. Section 6.2, describes the
parameters under which the results are discussed. The results for each ambiguity category
are then dealt with, in turn, in the subsequent five sections. In Section 6.8, the context
sentence results are juxtaposed with the ambiguous parts of the isolated sentences. The
conclusion brings this chapter to an end.
Chapter Seven: Analysis of Results
This chapter treats the conclusions that were drawn from the results of the experiment.
Section 7.2 deals with each category of ambiguity and compares the results obtained for
each of them. Section 7.3 discusses the results in relation to male and female participation.
Section 7.4 considers the linguists’ results and how they differ. Section 7.5 discusses how
bilinguals performed in the experiment and the processing theories of Chapter Three are
discussed further in Section 7.6. The final section of this chapter simply discusses the main
findings in light of the analysis.
Chapter Eight: Conclusion
This chapter brings together all the points that have been raised by the experiment and
subsequent results. It discusses the achievements of the experiment, the knowledge gained
and how this was possible. The paper finishes with a brief description of the difficulties
encountered and areas of possible future study.
14
Chapter 2
Temporary Syntactic Ambiguity
15
2.1
Introduction
Within the study of language, linguistics can be dissected into five main topics.
They are phonology, morphology, semantics, phonetics and syntax. Each one plays its own
role in the make up of language with some topics being more important in some languages
than in others. Syntax is universally one of the most important features of language and is
dealt with in this paper, together with semantics; the meaning of language. Both are closely
linked, for one cannot refer to the meaning of a sentence, until one has first processed the
constituent elements of that sentence2. In English, the main device for describing
relationships is in fact, word order itself. This implies that syntax plays a crucial role.
2.2
What is Syntax?
Languages have rules. The rules governing language are referred to collectively as
grammar. The reason for having rules is that people need to be able to speak an
indeterminate number of sentences in their lifetime. If each of these sentences had to be
learned separately, the task would be impossible. So, learning the rules for connecting
words, it is possible to create an infinite number of sentences, all of which are meaningful
to the people who understand the syntax. It is possible, therefore, to construct sentences that
the speaker or listener have never heard before. Thus, from a finite number of rules, an
infinite number of sentences can be simultaneously understood.
For this system to work with any degree of success, the rules have to be precise and have
to be consistently adhered to. The rules cover such things as: the way words are
constructed; the way the endings of words are changed depending on the context; the
classification of words into parts of speech (noun, verb, pronouns, etc ...) and; the way the
parts of speech are connected together.
2
There exists research, dealt with in chapter 2.3.2, to show that we may process semantic information after a
quick, unfinished syntactic analysis.
16
The rules of grammar do not have to be explicitly understood by the speaker or the listener
of the language. The majority of native speakers of a given language will have no formal
knowledge of the grammar of that language but are fully capable of speaking it to a degree
of grammatical accuracy. The rules are subconsciously assimilated whilst the language is
primarily being learned as a child. The syntax of a language is thus, the way word elements
must be construed to make well-formed utterances.
Syntax is an English word derived from the Latin syntaxis, meaning to arrange
together. There are two definitions, closely linked, that form the use of syntax today;
1.
2.
(a) The way in which linguistic elements (as words) are put together to form
constituents (as phrases or clauses)
(b) The part of grammar dealing with this
A connected or orderly system: harmonious arrangement of parts or elements2.
In order to clarify the definition of syntax, take the following sentence as an example:
subject
1.
verb
object
The boy
loves
his dog
This sentence follows the standard subject-verb-object word order. If the order of the
words were switched in some way, either the meaning of the sentence as a whole would
change or the sentence would be rendered meaningless. If one takes a transitive verb such
as to give as an example, a different syntax can change the meaning completely:
2.
The boy gave the dog the woman
3.
The boy gave the woman the dog
By merely changing the word order of the sentence, the whole interpretation of the
sentence changes. In (2) the dog is the direct object of the verb give i.e. its being given. In
(3) the object being given is the woman. One can see the great reliance on syntax in
English, at least, to disambiguate the meaning of sentences.
2
Definition gathered from Merriam-Webster Online Dictionary
17
This diagram shows a syntax tree analysis for a basic sample sentence like “The boy ate the bread”3
The tree diagram above illustrates the simple syntactic structure of a sentence. One can see
the hierarchy involved amongst its nodes. A sentence is broken down into a noun phrase
and a verb phrase. The noun phrase is acting as the subject of the sentence whilst the
function of the verb phrase is to predicate over the sentence. Each of these, in turn, can then
be broken down into their individual constituents until no further decomposition is possible.
The article, noun and verb in this instance are the smallest syntactic units of the sentence.
They correlate to individual words. For every grammatical sentence, there is a syntactic
structure like this that can describe it. The longer and more complex the sentence, the larger
and more complex the syntactic analysis will be.
2.3
The Role of Syntax in a Language
The understanding of syntax is an integral part of mastering any language. If one
does not grasp it during the early years of childhood, many language problems can ensue.
The controversial debate, whether the faculty of syntax in language is innate or learned,
divides many linguists throughout the world. Whether it is down to nature or nurture
however, the fact of its importance remains.
The role of syntax, as referred to above, may vary from language to language. In
English it is quite important, but in a language like Latin, syntax is much more flexible.
This can be attributed to the fact that word endings in Latin indicate the case of a noun or
an adjective; such inflections make it unnecessary to rely on word order to indicate a word’s
function in a sentence. The relationships in a sentence must be explicitly stated, some
languages do this through heavy reliance on syntax, others achieve this through reliance on
word cases. In all languages, however, it plays a role of sorting smaller units of meaning
into larger units of meaning, like sentences or phrases. The meanings of such sentences
often depend on the syntactic structure employed. The absence of an integral syntactic unit
can often reduce an utterance to nonsense:
3
This image was taken from a website, www.harmony.org.uk.
18
noun phrase
4.
noun phrase
The man
library
The syntactic rule, that to make a well formed sentence it must contain at least one noun
phrase and one verb phrase, is broken here. If a native speaker of English reads the sentence
(4), without any formal knowledge of the rule, they can instantly see that it is nonsense.
They may try to insert a predication over the sentence but if it originally lacks one, the true
meaning cannot be conveyed. Adherence to the syntactic rules of a language is
consequently a fundamental prerequisite in conveying a message.
In second language acquisition, many problems arise out of the fact that there is a different
syntactic structure to convey the same meaning within the target language. Take the
following sentences as an example of the difficulties faced:
subject
verb(auxiliary)
verb
pronoun
adverb
He
has
5.
done
it
already
This is in contrast to the French equivalent sentence:
subject
verb(auxiliary)
6.
pronoun
adverb
verb
Il
l’
a
déjà
fait
These examples illustrate that even the most basic sentences, under translation, can require
a syntactic structure that may be alien to a speaker of the source language. Whilst it comes
naturally to a native speaker, the importance of a correct syntactic structure can often be
vital for a foreigner to achieve understanding. Syntax can therefore be seen as the building
blocks required in creating meaningful utterances. The topic that will be discussed in the
next section is that in certain utterances, can you have two syntactic interpretations for the
same sentence?
2.4
Ambiguity
19
Ambiguity is defined as:
1.
Subjectively wavering of opinion; hesitation, doubt, uncertainty, as to one’s
course
2.
An uncertainty, a dubiety
3.
Objectively: Capability of being understood in two or more ways; double or
dubious signification, ambiguousness
4.
Word or phrase susceptible of more than one meaning; an equivocal expression4
A word, phrase, sentence or other form of communication is called ambiguous if it
can reasonably be interpreted in more than one way, as the above definition suggests. The
simplest example of this is the smallest semantic unit, a word5. Take an example such as the
word “row”. This word, taken in isolation, can refer to many things due to its homonym
status. It can mean the act of rowing a boat, an argument between two people6 or a queue of
people in a line. Sometimes this is not a serious problem, at the word level at least, as the
context in which it’s used usually disambiguates the intended meaning. For instance, if one
read, “they all formed into a row”, it is unlikely, though not impossible, that it means they
had begun arguing or that they were rowing a boat.
More problematic, however, are words whose senses express closely related concepts. Good
for example, can mean “useful”, “functional”, “exemplary”, “pleasing”, “moral” and
probably other similar things. If you said “I have a good daughter”, you would not know
which interpretation is intended.
There are several different types of ambiguities in language, some of which are used
purposely, whilst others cannot be avoided. In spoken language, lexical ambiguities can
arise as in the following example:
7.
8.
“Ice cream”
“I scream”
This is rarely a problem however because it is nearly always disambiguated by the context
in which it is uttered. Philosophers and other users of logic, spend a vast amount of time
removing ambiguities from arguments because it can lead to incorrect conclusions and can
equally conceal bad arguments. In the case of a politician, this can be used to good effect.
9.
“I oppose taxes which hinder economic growth”
If a politician announced the above statement (9), a number of interpretations could be
considered equally valid. In each case, the interpretation is expected to suit the way in
which the hearer desires it to be understood. Some, might conclude that the politician
4
Definition gathered from Oxford on-line dictionary, dictionary.oed.com
A morpheme is the smallest semantic unit but for the purpose of this paper a word will constitute such a
notion
6
This meaning is not strictly a homonym as its phonemic qualities are different to the other meanings. It
suffices as an example because it is orthographically the same.
5
20
opposes taxes in general because they hinder economic growth. On the other hand, there
will be those who interpret the statement as though the politician opposes only those taxes
that he/she believes will hinder economic growth. There are also other ambiguities and it is
the topic of syntactic ambiguity with which this paper is primarily concerned.
2.4.1
Syntactic Ambiguity
Syntactic ambiguity is a property of sentences that can be parsed in more than one
way. This may or may not involve one word having two parts of speech as in many
homonyms. Due to syntactic ambiguity, it is often difficult to understand the intended
meaning as in the next example:
Figure 2.3.1.1
S
VP
PP
NP
Det
The
N
NP
V
Det
NP
N
P
Det
girl
the
with
telescope
N
saw
man
the
Figure 2.3.1.1 describes a valid syntactic interpretation of a sentence. This
interpretation is depicted by the fact that the prepositional phrase is at the same level as the
noun phrase and verb. In other words, it implies that the prepositional phrase is modifying
the verb itself and in doing so renders the interpretation that, it is with the use of the
telescope that the girl was able to see the man. This is in contrast to figure 2.3.1.2.
21
Figure 2.3.1.2
S
VP
N’
PP
NP
NP
NP
Det
N
V
Det
The
N
P
Det
N
girl
the
with
telescope
saw
man
the
The interpretation here is somewhat different. The syntax tree in this instance
describes the situation in which the prepositional phrase is below the noun phrase, N’, and
at the level of the other noun phrase relating to “the man”. The interpretation that is
intended from parsing the sentence like this would imply that the girl simply saw the man,
who was carrying a telescope on his person.
This is just one example of how the intended meaning of a sentence can be
syntactically ambiguous. Depending on the position of the syntactic nodes, the meaning of
the sentence can change dramatically. In the above sentence however, there are only two
possible syntax structures. In the next example there can be four.
“Time flies like an arrow”
Although one immediately jumps to an interpretation that best fits a sentence like
this in isolation, there are many other ways that it could be interpreted, however bizarre.
The four possible meanings of such a sentence can only be told apart by their syntactic
analysis. They are all valid parses but it is the first interpretation that fits best and as a
result, is the one most likely to be true in an isolated reading.
a.
b.
c.
d.
Time passes as fast as an arrow flies
Measure the speed of flies as you would an arrow
Measure the speed of flies like an arrow would
A kind of fly, time flies, like arrows
There are hundreds of possible examples of syntactic ambiguity in every natural
language in the world. Our brains can decipher the meaning of a sentence in a split second
22
yet if there are two possible interpretations, it chooses one over the other. This must be
accomplished by other than purely syntactic means and will be dealt with later in the paper.
But first, one must look at the concept of temporary syntactic ambiguity.
2.4.2
Temporary Ambiguity
Temporary ambiguity is, as its name suggests, ambiguity that is temporary. In other
words, it is ambiguity that can be resolved by the individual constituents of the sentence,
but at a slower pace than if the sentence did not contain such ambiguity. It is also
commonly known under another guise, the Garden Path effect. The Garden Path effect is a
phenomenon that has been well researched, see Crain & Steedman (1985). The concept is
self descriptive as one is “led up the garden path” by the structure of the sentence and hence
difficulty arises when the first syntactic parsing does not fit the rest of the sentence. For the
remainder of the paper the term ‘temporary ambiguity’ will be used. Temporary ambiguity
usually arises if part of the sentence is compatible with two syntactic structures.
10.
While Nigel was reading the newspaper
Could be continued in at least two different ways:
11.
12.
While [Nigel was reading the newspaper] his tea was getting cold
While Nigel was reading [the newspaper lay unnoticed on the floor]
In (11), “the newspaper” is the direct object of the verb “read” whereas in (12), “the
newspaper” is the subject of the main verb “lay”. This means that when a listener or reader
comes into contact with a sentence that begins like that at (10), then it is impossible to be
certain which structural analysis is required to complete the sentence. It could be that of
(11) or (12) which implies that “the newspaper” may initially be incorrectly analyzed. This
is a classic example of temporary ambiguity.
One must concede that every type of ambiguity treated earlier in the paper can, in
fact, fall into the broader sense of temporary ambiguity. All that is required in each case is
that some pragmatic or semantic information be added in order to resolve the true meaning.
Temporary ambiguity in that sense, however, does not strictly follow the provided
definition. The discrepancy arises in that the sentences cannot be resolved by the individual
constituents of the sentence, i.e. further information is required. Temporary syntactic
ambiguity is the main topic of this paper and conforms to the narrower definition.
Temporary syntactic ambiguity is therefore ambiguous, but only for a short time and can be
resolved purely on the basis of the constituents of the sentence.
Figure 2.3.2.1
23
VP
N’
PP
PP
NP
V
Put
Det
NP
N
P
Det
NP
N
P
Det
N
the
on
napkin
the
frog
the
in
box
Above, is an example of temporary syntactic ambiguity in a sentence. It is a fully
grammatical sentence. As it illustrates, there is only one possible way to parse such a
sentence and in that sense, it is syntactically unambiguous. Yet it falls into the category of
temporary syntactic ambiguity, due to the way in which we process the information that we
gather from the sentence. The unfamiliar syntactic construction of two prepositional phrases
side by side in a sentence, demonstrates the difficulties faced during processing. Whilst
these types of sentences can be described as temporarily syntactically ambiguous for this
reason, they cannot be referred to as purely syntactically ambiguous.
The illustration in figure 2.3.2.1 is just one example of temporary ambiguity. There
are several other syntactic constructions that create the same kind of temporary ambiguity
during processing. They will be dealt with extensively, later in this paper.
2.5
Temporary Syntactic Ambiguity in Language
As one might imagine, ambiguity is avoided in language wherever possible. This is
due to the fact that the main aim of producing an utterance in the first place, is to be
understood. Thus, syntactic ambiguity is generally avoided, making the occasions when one
might have to decipher such a message quite seldom. However, when one encounters a
message containing this type of ambiguity, other semantic or pragmatic factors are used to
decipher the intended meaning.
In relation to temporary syntactic ambiguity, it is not necessary to take any other
outside factors into account. It is therefore a more common occurrence than normal
ambiguity in language because one does not unconsciously strive to avoid it. The sentences’
24
constituents relay all the information that is required to interpret the meaning. If the
processing and reprocessing of the utterance does not result in a satisfactory outcome then
other pragmatic means may have to be employed. To fully understand temporary syntactic
ambiguity, we must explore at the thought processes that prevail in order to decipher such
an utterance.
Chapter 3
Sentence Processing
25
3.1 Introduction
Sentence processing is the term used to signify the process which our brains
undergo in order to comprehend an utterance. The earlier analogy to a prospector is quite
intuitive in that there are many current theories that draw on the notion of subconsciously
picking away at a sentence to gain its meaning. This paper is primarily concerned with the
processes involved in deciphering written material, i.e. sentence comprehension with regard
to reading.
Sentence comprehension is the ultimate goal of sentence processing. If someone
produces an utterance, it would be for the purpose of being understood by someone else.
Effective comprehension relies primarily on vocabulary and in order to understand, the
reader must possess the ability to decipher the basic lexical items that make up the sentence.
Word recognition can only be achieved through the concept of decoding. Decoding is
where the individual letters of a word can be correlated to phonemes, thanks to alphabetic
principles, which can be blended together to allow a person to recognize a word.
Word recognition can however, only help the reader to a point. Sentence
comprehension involves much more than just recognizing the words of a sentence. To
become a successful reader, one must also be able to comprehend the words read, within the
context of the sentences containing those words. In other words, one must weave the
meanings of the individual words into the meaning of the sentence.
26
If one were to choose a tent as another analogy, one could imagine being faced with
the task of taking down a tent that one has never seen before. Granted, most tents are quite
easy to take down, as are some sentences, but what would happen if it were an antique tent?
It is conceivable, that one may have some difficulty in packing it all away. In similar
fashion, sentence comprehension relies heavily on the person’s ability to decipher the
structure of a sentence, i.e. the syntax. In addition to this, the reader must be able to retain
all of the words that have been read, until the entire sentence has been processed. Such
storing of words is commonly termed, working memory. As syntax plays such an important
role in sentence processing, it is an issue that we will look at first.
3.2 The Role of Syntax in Sentence Processing
Syntax is an important part of sentence processing. Consider the following simple
sentence:
1.
Mary saw a cat
The main process being employed is that of word association. In sentences like this,
syntax plays a secondary role and is only employed to corroborate the associations. When
the associations are ambiguous or the syntax is complex, however, simple word
associations will not provide a satisfactory understanding. One must have a clear
comprehension of syntax in order to establish how each word forms part of the sentence
structure and subsequently understand the sentence in its entirety.
It is obvious that when a sentence is encountered, one does not casually puzzle
one’s way through in order to achieve an interpretation. Sentence processing is, in fact,
completed at very high speed and we are largely unaware of the processes involved. We are
aware of its presence, however, as is evident by the final outcome; comprehension.
The role of syntax in sentence processing, as elaborated, still remains rather vague.
This is due, in part, to the fact that complete agreement does not exist between scientists, as
to the exact function played by syntax in sentence processing. In Sections 3.5.1 and 3.5.2,
syntax will be considered again, but in relation to each of the hypothesised processing
theories. It may then be possible, to define the role of syntax more accurately within the
context of sentence processing. Nevertheless, it is worth mentioning, that in each case,
syntax and syntactic structures play a vital role in the ultimate comprehension of a sentence.
3.3 The Syntactic Challenges in Processing Written
Sentences
27
Children are constantly learning about syntactic structures without being unduly aware of
it and learning to listen and speak are the primary sources of this knowledge. By the time
they start learning to read, their considerable knowledge of syntax can be employed to
manipulate that which they are reading, so that a correct interpretation can be obtained.
There are a few relatively simple strategies that a learner employs in order to process a
sentence and this section will discuss the concepts that are relevant to the processing of
sentences, which are learnt at an early age.
3.3.1
Word Order
In English sentences, the word order is generally in the form of subject-verbobject7. As a result, a learner encountering a new sentence will try to apply this order to the
corresponding words in the sentence. This is, of course, assuming that that each sentence is
processed by structure first and meaning second.
This very point will be picked up later in the section but for now, it is suffice to say,
that due to reliance on the SVO word ordering, early readers can understand simple active
sentences such as:
2.
He called her
But have much more difficulty in correctly parsing a sentence in the passive voice
such as:
3.
He was called by her
Often a passive sentence such as (3), is interpreted as ‘he called her’ so that it
complies with the ordering that was previously learned by the young reader.
3.3.2
Minimum-distance Principle
Another learning reader tendency is that of minimum-distance. This terms refers to
the incorrect assumption that a word refers to the closest related word. For instance, they
might assume that a verb refers to the closest preceding noun or, that a reflexive pronoun
refers to the closest preceding noun of matching gender. Take the following as an example:
4.
She brushed her hair all by herself
5.
She brushed the hair of the short girl all by herself
7
SOV and SVO orderings are the most common in languages but all permutations do exist in some form
28
In the first sentence, a young reader would have little problem matching up the
reflexive pronoun ‘herself’ with the correct reference pronoun, ‘she’. On the other hand, the
second sentence would cause greater difficulty as the reader would try to match the
reflexive pronoun with the noun ‘the short girl’. Readers usually tend to rely on the
minimum-distance principle, when their working memories are taxed by the need to
remember many words throughout the course of a sentence.
3.3.3
Coined-clause Analysis
Coined-clause analysis is a term used to describe the assumption by a young reader,
that the existence of two clauses in a sentence means that the clauses are automatically
conjoined. In other words, that there are two main clauses and that they are joined together
by a conjunction such as ‘and’. A classic example of this would be:
6.
The cat caught the mouse that chased the shrew
7.
The cat caught the mouse, and the cat chased the shrew
The first sentence is the original sentence containing the relative clause. The second
sentence is a common misinterpretation of that sentence. A beginning reader would often
make such a mistake whilst trying to apply a syntactic analysis.
The three strategies described above would be used by a naïve reader to fall back on
in the event of processing an unknown sentence. They have not yet read extensively enough
and therefore have only enough syntactic ability as their spoken tuition has allowed. Their
sentence processing ability is constantly improving with practice and along with this
maturity, the increase in efficiency of their working memory. Soon they are capable of
switching from these strategies to paying attention and comprehending the actual syntax of
the sentence. However, they are important to mention as matured readers, in times of
difficulty, will naturally fall back on techniques such as the ones discussed. There are other
syntactic structures as well that pose problems for a beginner. They include the passive
voice, negation and embedded clauses. The difficulties with these will now be described.
3.3.4
Passive Voice
Passive structures are harder to understand, learn and produce than their active
counterparts. This is due to the object-verb-subject (OVS) construction of the sentence.
Passives are even harder to interpret if they are reversible. Take the next couple of
sentences as an example:
8.
He was called by her
9.
The house was burgled by the thief
29
The first example shows the difficulties faced. In this case either the guy or girl
involved could technically be able to do the calling. In the second example however, the
house could hardly be capable of burgling anything and so a correct interpretation would
result relatively quickly. It takes time and practice to refine such processing techniques so
as to correctly interpret a passive sentence like in (8) without any difficulty. If passive voice
requires such difficulty to process and understand then why do we use it? Would it not
make life easier to forget about it entirely? The short answer is no. Passives are useful for
many reasons. They are useful, for instance, when you really don’t know who the subject
is:
10.
My wallet has been stolen
Or in order to emphasise something in the sentence:
11.
It was the women that was seen by the old man
Or even when a non-active noun has to be used as in the sentence:
12.
A non-active noun must be discussed!
The majority of sentences are in the active voice but passive voice is the next most
used structure in written text. It is for this reason that one must grasp the structural analysis
required for the passive voice as early as possible.
3.3.5
Negation
Sentences in the negative are also quite challenging for a young reader to master.
When reading a sentence containing negation, the reader generally processes the positive
sense of the sentence and then takes another processing step to arrive at the negative sense.
For instance, a learning reader will find it easier to process and respond to a sentence such
as:
13.
Can you see an object that is square and blue?
In comparison to a sentence such as:
14.
Can you see an object that is square and not blue?
Once again, practicing and contact with negative sentences improves
comprehension and reading fluency so that a person can be relatively comfortable with
constructs like this quite quickly.
30
3.3.6
Embedding
Embedded sentences can prove quite difficult to process. As already described
above in the coined-clause analysis, young readers find it difficult to correctly interpret a
sentence in which the object of the main clause can be the subject of the embedded clause
like:
15.
The lion bit the buffalo that chased the hyena
This is in contrast to the ease at which one can process a sentence in which the
subject of the main clause is also the subject of the embedded clause as in the following
example:
16.
The lion that chased the buffalo bit the hyena
There are even more examples of where embedded clauses cause learning readers
more difficulties. For instance, sometimes the subject of the main clause is the object of the
embedded clause:
17.
The lion that the buffalo chased bit the hyena
These types of sentences are particularly difficult to process as they violate not only
the minimum distance principle but also the subject-verb-object sentence word order
structure. They will be seen again throughout this paper as a difficulty that can cause
problems to even an accomplished reader.
Another type of embedding that causes processing difficulties, especially in the
naïve reader, is that of cleft transformation. This type of embedding changes the subject and
object order just like the passive voice. Cleft transformation increases the focus on a
concept by dividing a perfectly serviceable single clause into two clauses, each with its own
verb.
18.
The girl drank the juice
19.
It was the juice that the girl drank
The sentence at (18) is a perfectly good sentence. If one needed to emphasise the
juice, then the cleft transformation is required as in (19). Early readers soon pick up all
these processing quirks and have refined an efficient technique within a couple of years that
stand to them for the rest of their lives. The more they come in contact with structures as
described in this section, the quicker they learn to think carefully and logically about
sentence structure. Ultimately they arrive at processing techniques that are fully matured
and reasonably flexible to the consumption of new, unseen sentences.
31
3.4
The Role of Working Memory in Sentence Comprehension
Just like syntax, the working memory plays a vital role during the processing of a
sentence. Working memory is the conceptual part of the brain that we employ to store the
words of the sentence that we have already read. Syntactic comprehension can only achieve
so much if we are unable to remember the words that make up the sentence.
When one encounters a sentence, each word is processed as fully as possible the
first time it is read. The results of processing each word such as the letters, the sounds, the
meaning, and the word associations are then all stored in working memory until the entire
clause has been read. Different clauses contain a different amount of words, some large,
some small, thus efficient storage can prove quite important in the overall retention of the
clause. For instance:
20.
My new shoes, my red book, my pencils and my papers were lying on top of each
other on the table
This sentence is relatively simple to process but it contains a large amount of words.
Efficient storage is the means that we rely on to remember all the details of such an
utterance.
As someone recovers a clause from their working memory, they form the gist of
what is implied by the message. In other words, they form an interpretation and mental
representation of the clauses meaning. This is then kept in working memory until
subsequent sentences, clauses or phrases are read to modify this into another interpretation
of the intended meaning. Naturally enough, the process becomes complex when one clause
contains a reference to the subject or object of another clause. The process is even more
complicated when an entire clause functions as part of another clause as in the following
sentence:
21.
The fact that he couldn’t finish the race doesn’t faze me in the slightest
As one reads down through a passage, a hierarchical structure of each gist is
constructed in working memory. For example, a sentence is processed which leads to an
interpretation for that sentence. A subsequent sentence or clause may modify this
interpretation slightly. A higher level gist is then constructed to encompass everything that
has been read. Eventually, one interpretation succeeds another until the whole paragraph
has been read thus creating a gist for the entire paragraph. A person’s ability to temporarily
store words in working memory depends on many factors such as their age, experience and
language proficiency.
The code that people use to store this information in working memory is always
phonological. For this reason, phonological decoding skills are crucial for the use and
development of working memory abilities. Therefore, someone with efficient decoding
skills will possess the ability to create and maintain effective phonetic representations of
written words and will be able to retain, in working memory, both the words themselves
and the order in which they occur in the sentence. An efficient working memory is an
32
advantage in understanding not only the sentence at hand, but also the paragraph, the
chapter, the book, the series and so on. The longer the passages one reads and understands,
the greater the development of the working memory and hence the greater the ability in the
future.
3.5
The Processing Theories
As one would imagine, it is very difficult to find agreement among scientists as to
how we actually process a sentence. This is due to the fact that we are still largely unaware
of the neural processes that take place whilst processing an utterance. There have been
many conflicting experiments that favour one theory in the place of another. Nowadays,
there are two main theories that seem to be the most plausible models to most scientists,
serial processing and parallel processing. They are, however, completely different from
each other with regards to the way in which each part of the process is activated in the brain
and the interaction among the individual parts during the processing.
3.5.1 The Serial Model
The serial model is, as its name might suggest, a linear chain of neurological
processes that link with one another to arrive at overall comprehension. Each part of the
processes involved do not interact with each other, they only pass information on to the
next process in a manner described by the illustration below:
Figure 3.5.1.1
Phonology
Lexicon
First-pass
Parsing
Syntax
(Second-pass parsing)
Reanalysis
Semantics
Repair
In the first stage, a phonological representation of the sentence is processed. This is
then handed on to the lexicon facility that recognises the phonological representations as
words. These words are retrieved and passed on to the syntax faculty in the brain that goes
through three sub-stages. The first is a first-pass parsing action that filters the words into an
initial syntactic structure. This is then handed on to a reanalysis phase that makes sure all
the constituents fall correctly into a satisfactory parse.
It is worth mentioning that at this stage, according to Frazier&Fodor (1978), the
semantic context first plays a role. The conclusion to their experiments favoured a process
33
in the reanalysis phase that involved the evoking of a rough semantic representation during
which a meaning is mapped to the constituents of the sentence. The syntactic analysis
would then reconvene in a repair stage where the final output is computed.
Other scientists disagree with this and favour a process where the output of the
syntactic process is further refined by the repair stage before being passed on to the
semantic analysis phase. Holmes (1979) states that:
“The syntactic approach…requires that the subject complete the syntactic analysis of at
least a clause before semantic analysis can begin”
If the serial model is the correct model of the human ability to process sentences
then each link in the chain corresponds to a separate and non-interactive step in the process
to comprehension. An overall diagram of the language processor is suggested by Forster
and is illustrated in figure 3.5.1.2. The serial model is the basis for such a processor.
Figure 3.5.1.2
Syntactic
Processing
GPS
Lexical
Processing
Input Features
General Conceptual Knowledge
Message
Processing
Decision output
Organisation of language processor8
According to this structure of the language processor, the input features would be
the individual words being read. The GPS (general problem solver) has no linguistic stimuli
and is employed to process any problems. For the purpose of comprehension however, it
needs linguistic input and it receives this from the three processors.
8
Illustration taken from Levels of processing and the structure of the language processor
Forster, K.I., (1979)
34
The lexical processor sorts the inputs and hands the results to the syntactic processor and
the GPS.
The syntactic processor sorts the results into a syntactic structure and passes the results to
the message processor and the GPS.
The message processor takes the syntactic analysis and maps a semantic representation to
it before handing the results to the GPS.
The GPS gathers all this information from the processors and together with the help
of the general conceptual knowledge of the surrounding world, the GPS can judge if it is
a satisfactory message analysis and output a decision. The decision reached can be seen as a
correct interpretation of the message and a subsequent decision is made.
The serial model is a plausible way that our language faculty in the brain processes
a sentence. It has, in recent times, come under a lot of scrutiny from scientists. Many
experiments have been conducted that find flaws with this model in favour of the model in
the following section, the parallel model.
3.5.2 The Parallel Model
The parallel model differs from the serial model in one main way, the influence the
semantic mappings have on the other processes. Whereas the serial model had each stage
independent of one another, its parallel counterpart allows multiple interactions between all
stages of the process. In other words, all sources of information are available
simultaneously. This semantic approach is described by Holmes (1979) as:
“scanning the input for the major content words and putting these together in the most
sensible way. Purely syntactic analysis, such as paying attention to the order of words,
their inflections, and so on, is only necessary when this preliminary semantic analysis fails
to come up with an unequivocal message.”
Many experiments have employed ambiguity resolution as a means to conclude that
the parallel model is the only theory that stands strong in the face of criticism. It concedes
that there must be an initial syntactic “surface structure” analysis but that this is directly
followed by a semantic analysis. Any deep-structure syntactic analyses are left until after
the two previous stages have been completed. The following diagram illustrates this.
Figure 3.5.2.1
35
Surface structure analysis
Semantic analysis
Check accuracy of
deep structure
relations
The parallel model is also a plausible explanation of the way we process a sentence.
It simply advocates that both semantic and lexical information are used in the process of
syntactic analysis. Experiments have showed that it is likely that we, in some way, map
semantic associations before we finish processing the sentence. Using the parallel model as
a basis, Forster (1979) suggests a system of processing language, in figure 3.5.2.4, that
vastly differs from the processor based system suggested for the serial model.
Figure 3.5.2.2
General
Conceptual
Knowledge
Inference
Rules
GPS
Lexical
Information
Syntactic
Rules
Decision
Output
Semantic
Rules
Language processing system without language processor9
In this diagram, the GPS collects information from all areas of the language
faculties. These messages can be intertwined allowing for a process in which results so far
can be relayed and used by other areas to guide a correct interpretation. The final decision
output is reached by the interaction of all these processes and the overall judgement of the
GPS. In this way, a brief structural analysis can be passed to the GPS which then uses
semantic rules to find a possible interpretation. If a deep structure analysis is required, the
information can then be passed back to the syntactic analyser again.
9
Diagram taken from Levels of processing and the structure of the language processor
Forster, K.I., (1979)
36
The two models presented show that it is very difficult to totally disprove one model
in favour of the other. Experiment results show varying biases towards one or the other
model. In his conclusion, Holmes (1979), states:
“The evidence I have presented amply demonstrates that any model of sentence
comprehension that minimises the importance of syntax is misguided”
He continues by offering another conclusion in light of his experiments with
ambiguity resolution:
“Semantic analysis is preceded by a superficial analysis in which a unique surface
structure representation is imposed on the clause”
Further research is required if we are to find out which of the two models better fits
our own ability to process sentences. Later in the paper I will refer to these two models
again and will offer, under the guidance of my own results, which model seems the most
likely to be used by humans.
37
Chapter 4
Methodology
4.1
Introduction
38
Syntactic and semantic analyses, as have been discussed in the previous two
chapters, combine in some neurological manner during the processing of a sentence to
produce what we know as comprehension. In the last section, two theories were put forward
to explain this phenomenon. To gain a greater understanding of which process is a more
likely model of the way we do, in fact, comprehend, an experiment has been compiled.
Drawing from research documented in the first section of this paper in relation to temporary
syntactic ambiguity, the rest of this paper will be concerned with discovering more about
how we process such analytical pitfalls. More precisely, how much longer it may take us to
resolve a sentence in which temporary syntactic ambiguity is present and also, which type
of temporary ambiguity we find most difficult.
This type of study naturally raises the question of which people can resolve this
ambiguity more quickly and also whether this is an advantage to some people? The
domains that this paper will be concerned with are the traits of a bilingual’s processing of
temporary ambiguity. It is also concerned with whether or not contact with linguistic
instruction prior in one’s life helps to resolve these difficulties more quickly. Each of these
questions will be dealt with in chapter seven when the results of the experiment are
discussed.
To study the way in which we process a sentence is extremely difficult. After all,
we are still largely unaware of the neurological processes involved. However, there are a
few areas that we can analyse. The most obvious of these is the time factor. With this in
mind, section 4.2 attempts to document the parameters that must be adhered to in order to
create a valid experiment. These parameters are described from research in the area. The
experiment that has been undertaken in this paper is a psycholinguistic study of how
quickly or slowly we can resolve temporary syntactic ambiguity. In achieving this, five
different categories of temporary syntactic ambiguity have been used. These categories will
be explained further in the section 4.3. In each case, a piece of software was used,
ICECUP3, to traverse the ICE-GB corpus of English to find examples of each category of
temporary ambiguity. This is further described in section 4.4. This will lead on to the final
section in the chapter where the final design of the experiment will be discussed.
4.2
Experiment Parameters
To conduct any experiment, a set of rules must be adhered to in order to maximise
the validity of the conclusions that can be drawn from it. In his book, “The Empirical Base
of Linguistics”, Schütze assesses traps that other linguists have fallen into. He describes
possible biases and general issues that can arise whilst constructing an experiment and also
advises the experimenter on how to avoid such obstacles. According to Schütze (1996),
these factors can be split into two main groups; materials used i.e. the sentences employed
in the experiment, and the procedures employed, which include the subjects, the
instructions, and pitfalls to avoid and the second. The first of these groups, the procedures,
39
will be treated in the subsequent subsections. The materials used will be discussed in the
next section.
4.2.1
Selection of Subjects
The selection of subjects is an obvious starting point when devising an experiment.
The correct strategy of choosing participants is vital in validating an experiment’s
conclusions. The subjects must be a diverse cross section of the population in order to claim
that they are a representative group of the entire population, and that their judgments are
representative of the entire population. For an experiment of the kind this paper describes,
such a prerequisite is necessary. Schütze writes:
“…that if it is the study of competence of normal native speakers that we claim to be
investigating, we need to study random samples of normal native speakers.”
In relation to the current experiment, the average age and sex of the participants
must be kept track of. This is a parameter that must be followed in order to validate a
representative group of the entire population. How this is achieved will be discussed in
section 5.4. Schütze feels that this is a parameter that theoretical linguists rarely adhere to.
Another pitfall that Schütze warns against is that participants must not be people
with any linguistic training. As one might expect, through the very nature of studying
language, a linguist might unconsciously be applying rules that a non-linguist would not
even be aware of. Greenbaum (1977b) states that a linguist:
“…has been examining over a period of time a set of examples exhibiting close
similarities and therefore his judgments will tend to become blurred”
and also that:
“he is inevitably prejudiced by his general theoretical position and by the specific
hypothesis he is testing”
In light of these prejudices, the linguistic background of the subjects in an
experiment must be discovered so that valid conclusions can be extracted from the exercise.
In relation to the current experiment, the requirement that the subjects not be a linguist is
not strictly adhered to however. This is due to a couple of reasons. The first is that the
experiment, by definition, includes a comparison that can be made between linguists and
non-linguists. It can be conceded that linguists may have an upper hand at tackling
temporarily ambiguous sentences. But, instead of these subjects being omitted from the
experiment, they have been included in order to compare how much of an advantage they
may possess. To achieve this, the subject’s linguistic background must be documented after
completion of the experiment and compared later in the results. This will be discussed as
the topic of section 7.5.
40
4.2.2
The Instructions
The instructions are also a crucial part to the experiment. There are several
components to take into account when drawing up the instructions for an experiment.
Schütze recommends that linguists must make specific exactly what is required of the
subject. A definition or explanation of any of the terms that are required in the experiment
should also be present. Schütze explains that instructions “…must be specific…”
Another important parameter that must be acknowledged is the necessity of keeping
the statement of instructions to a reasonable length. If this is not the case, obvious
inaccuracies in relation to results would surface. The participant’s concentration is the key
to accurate results and if the length of instructions is too large, they may get bored and not
read them properly.
Greenbaum (1977b) suggests telling the participants what the experiment is really
about so that they will not introduce variability in the results by making guesses as to what
it’s about. If you don’t explain to subjects what you want, each one takes his own
interpretation and the results may become meaningless. This parameter is not adhered to in
the instructions of the current experiment for a few reasons. They will be more fully
explained in the section 4.5.2 when the design of the experiment is being discussed. For
now it suffices to mention that the less the subject knows as to what the experiment is
about, the more accurately one can obtain results for the way in which they resolve
temporary ambiguity. After all, the resolution is a largely unconscious process that should
remain so. The next section specifies how the material was found for the experiment.
4.3 Temporary Syntactic Ambiguity Categories
There are many temporary syntactic ambiguity categories. Many are commonly
known under the guise of garden path sentences. Five such categories were chosen to be
dealt with in this paper. They are:
a.
b.
c.
d.
e.
Complement clause Vs Relative clause
Object noun phrase Vs Main clause subject
Direct object Vs Embedded subject
Prepositional phrase attachment
Main clause Vs Complement clause
These five categories were chosen on the basis that they were well documented and
there were numerous examples that could be employed in the experiment itself.
Each category can be relatively easy to understand and this was another prerequisite
if bilinguals are taking part in the experiment. The easier the construction of the sentence,
the more likely the results will reflect the ambiguity resolution that is being measured.
41
In truth, any number of categories could have been chosen but five of the most
frequent ones were used. This can give us an overall view of which temporary ambiguity
might be the most challenging for either native speakers or bilinguals to process. This may
ultimately help future teaching methods to avoid such difficulties before they are
encountered.
4.3.1 Complement Clause Vs Relative Clause
The first of the ambiguities that are being dealt with in the experiment is between
the complement clause of a sentence and the relative clause of a sentence. The complement
of a sentence is traditionally a constituent of a clause such as a noun phrase or an adjective
phrase that is used to predicate a description of the subject or object of the clause10. There
are two possible functions that complements can play in a sentence, namely as an object or
subject complement.
1.
The nation made Berti the king
2.
Berti was the king
In the first sentence, the noun phrase “the king” has the function of object
complement. This can easily be resolved by asking the question, what did the verb in the
sentence do to the object of the sentence? The answer in this example can be clearly seen
as, “making Berti the king”. In the second sentence however, the complement is qualifying
the subject and is hence, a subject complement. A common question that can be asked to
verify this is, what did the verb do to the subject of the sentence? For this example, the
answer can be confidently given as, “being the king”.
In light of this, a complement clause can be defined as a notional sentence or
predication that is an argument of a predicate11. In other words, it acts in the same way as a
complement of a sentence except it has a clause status. Therefore it acts as the complement
of the hierarchically superior clause. Here are a couple of example sentences:
3.
We thought that you were coming
4.
For you to come would be a mistake
As you can deduce from the above examples, complement clauses can function as a
subject complement clause or an object complement clause. Sentence 3 acts as an object
complement clause as it answers the question as to what the verb does to the object, namely
the fact that you were coming. On the other hand, as a subject complement clause, sentence
4 states that “for you to come” is qualifying the subject i.e. what would be a mistake.
10
11
Definition taken from What is a complement?, www.sil.org
Definition taken from What is a complement clause?, www.sil.org
42
Complement clauses can become very complex and difficult to resolve when the qualifying
complement clause is embedded or nested; this is out of the scope of this paper.
A relative clause is a clause which describes the referent of a head noun or pronoun.
It often restricts the reference of the head noun or pronoun12. There are several types of
relative clauses such as external, internal, restrictive, non-restrictive, etc… The study of
these is outside the scope of this paper and will not be dealt with any further. There are
however, many different ways of spotting relative clauses but the easiest are by looking for
key words such as who or that. They can come in many guises such as the following two
sentences:
5.
The man who went to the shop
6.
The passengers leaving on flight fr045 are lost
The examples illustrate that despite being able to spot a relative pronoun, this
doesn’t always mean success in spotting a relative clause as sentence 6 contains no such
pointers. The question that can often be asked to verify the existence of a relative clause is,
does it qualify the noun that precedes it? In both the above cases this is a positive outcome.
Using the two areas that have been discussed above, the question must be posed as
to how these two concepts can be linked to temporary syntactic ambiguity. The answer is in
the way they are processed. Read the sentence below:
7.
The president told the journalist that the first lady annoyed that he should go jump
in the Potomac13.
On first reading the sentence, you might find that you paused, however little, at a
certain point in the sentence. This point in the sentence is where temporary ambiguity has
arisen and your brain needs a split second to resolve it. You are largely unaware of the
process. It happens as a result of the way we process the sentence. As we read through the
sentence, a syntactic analysis is being formed. When we stumble upon a word that does not
fit in to our syntactic analysis thus far, a reanalysis has to be employed. This is what is
responsible for the slight pause. To illustrate the processes involved in the previous
sentence, it is broken up slightly. After reading:
8.
The president told the journalist that the first lady annoyed
The syntactic analysis is expecting a word such as “him” or “the man” or some sort
of noun to fit the overall sentence analysis. Instead, when the word processed is “that”, a
reanalysis of the sentence is required.
The complement clause versus relative clause ambiguity arises in that we process
the sentence expecting “that the first lady annoyed” to be a complement clause. This is, in
fact, not the case as we should be reading it as a relative clause. It takes us a split second to
realise this and continue.
12
13
Definition taken from What is a relative clause?, www.sil.org
Sentence taken from Sentence comprehension, www.ling.udel.edu
43
4.3.2 Object Noun Phrase Vs Main Clause Subject
The second ambiguity category that the experiment contains is that of object noun
phrase versus main clause subject. These terms need less of an introduction to understand.
An object noun phrase is simply the noun phrase part of the constituent of the sentence that
functions as the object in the sentence. This could range from “the man” to “the ugly
duckling” as long as it acts as the object of the sentence. For instance:
9.
He added yet another comment
In this sentence, “yet another comment” acts as the object or equally, the object
noun phrase.
The main clause subject of a sentence is the constituent of a sentence that functions
as the subject of the verb in the main clause. It is commonly referred to as simply the
subject of a sentence. It is merely termed as it is in order to distinguish it from other clause
subjects.
The temporary ambiguity that may arise between these two concepts is extremely
slight and but can cause all sorts of semantic pitfalls. Take the following as an example:
While Wallace was reading the newspaper was burning14
10.
To illustrate what happens during the processing of this sentence, two diagrams
have been drawn out. Figure 4.2.2.1 shows a diagram of our syntactic analysis up to a
certain point in the sentence. After this point, a reanalysis must be performed as the
following word does not fit into the current syntactic analysis.
Figure 4.3.2.1
S
VP
Ad
NP
NP
V
Ad
PN
Pre
CP
Det
N
While Wallace was reading the newspaper ……was?
14
Sentence taken from Sentence comprehension, www.ling.udel.edu
44
So far, this is a quite plausible syntactic representation of the sentence. The
processing of the sentence draws us to a syntactic construction where “the newspaper” is
regarded as being the object noun phrase of the predicate. The next expected word might
therefore be a noun phrase such as “the leg” in “…the leg of the chair broke” or “he” as in
“…he slipped into unconsciousness”. Instead of this, we next process the word “was”. This
is neither expected nor desired in the current syntactic interpretation and therefore we must
undergo a reanalysis. After finishing the sentence we have resolved the difficulty and
analysed the sentence correctly as:
Figure 4.3.2.2
S
Adv
Ad
NP
S’
VP
NP
VP
V
Ad
PN
While Wallace
V
Pre
CP
was
reading
Det
N
Pre
the newspaper was
CP
burning
As you can see from the diagram, the syntactic analysis surrounding the underlined
region has changed dramatically from the first diagram. This is the area that was reanalysed
after the temporary ambiguity had arisen. The “newspaper” was correctly interpreted on the
second attempt to belong to the subject of the main clause and not the object noun phrase as
initially perceived.
4.3.3 Direct Object Vs Embedded Subject
The next section deals with direct object versus embedded subject ambiguity. The
terms require a little introduction before their ambiguous interaction can be explained. The
direct object of a sentence is a grammatical relation that exhibits a combination of certain
independent syntactic properties, such as the following:
•
The usual grammatical characteristics of the patient of typically transitive verbs
•
A particular case marking
•
A particular clause position
45
•
The conditioning of an agreement affix on the verb
•
The capability of becoming the clause subject in passivization
•
The capability of reflexivization15
This is a convoluted definition that requires a certain amount of background
linguistic knowledge to decipher. A direct object can however, be spotted quite quickly if
you follow a simple question, what does the verb do? For instance, if the sentence is “They
make cars for a living”, the question would be, they make what? In every instance, it is the
cars that they make and so the direct object is located.
An embedded subject is somewhat more difficult to find. It is however quite easy to
explain. The subject constituent of a certain sentence is merely embedded in the sentence to
the extent that it might cover a whole clause in the sentence. The subject clause is still
treated as the subject of the sentence, yet it may contain many elements that themselves are
good candidates for being the subject in their own right. Take the following as an example:
11.
The gardener sweeping the pathway which had become over run is too old
The underlined words in the above sentence altogether constitute the subject of the
sentence. It is made up of many words though. As you can see, this embedded subject
concept can be extended indefinitely creating additional, even more complex, sentences.
Combining these two linguistic phenomena together in a single sentence can result
in temporary ambiguity. The ambiguity in this case is raised when an embedded subject is
misinterpreted on the first parse as being a direct object. For instance:
12.
The smart student remembered the answer was probably misleading16
Here the reader would progress through the input until the word “was” is read. At
this point, the syntax of a full sentence has been satisfied in that the noun phrase, “the smart
student” and the verb phrase, “remembered the answer”, combine correctly. The noun
phrase “the answer” is therefore parsed as the direct object of the verb.
15
16
Definition taken from What is a direct object?, www.sil.org
Sentence taken from Sentence comprehension, www.ling.udel.edu
46
When the next word after “answer” is discovered, namely “was”, a reanalysis must
be performed so that the new structure can be accommodated. This may come in the form
of just a split second and once again, we are largely unaware of the process. This type of
ambiguity should not pose many problems to resolve. The possible reasons for this will be
discussed in chapter seven. For now, it suffices to mention that it is not very syntactically
challenging to reanalyse.
4.3.4 Prepositional Phrase Attachment
To discuss prepositional phrase attachment temporary ambiguity we must first give
a definition of what a preposition is. Quite simply, a preposition is a word that we use to
indicate location. Usually these prepositions show or relate this location in the physical
world. To define them more precisely, they are words or phrases placed typically before a
substantive and indicating the relation of that substantive to a verb, an adjective, or another
substantive17. There are more than one hundred prepositions in English which is quite small
in comparison to nouns and verbs but their relative frequency in the language is the
greatest. Some common prepositions are:
At
To
From
For
On
Over
By
With
Along
Past
Outside
Under
A prepositional phrase is therefore a phrase that contains a preposition as its head
and a modifying clause thereafter. Examples of such phrases are:
Figure 4.3.4.1
PP
NP
P
13.
14.
17
Det
N
On
bus
To
car
the
the
Definition taken from Oxford English Dictionary-online, www.oed.com
47
15.
Under
boat
the
At first sight you would wonder how these may be ambiguous in any way. The
simple answer is when they fall, one after the other, in a sentence, their attachment to
referent objects becomes temporarily unclear. For example:
16.
The cook placed the cake in the oven on the table18
Similarly to the last category of ambiguity, a fully grammatical sentence lies within
the main sentence. In this case, the syntactically sound utterance:
17.
The cook placed the cake in the oven
Is parsed in such a way that no more words are expected by the reader.
As the input is processed, the next word discovered, “on”, requires a reanalysis of
the syntactic structure up to this point as it does not fit the current “provisional” structure.
This reanalysis is caused by the temporary ambiguity of the prepositional phrase
attachment. In other words, on first reading, the reader assumes that the cook merely
“placed the cake in the oven”, but after the syntactic reanalysis, realises that the cook placed
the cake “on the table”. The prepositional phrase that was first thought to refer to where the
cake was placed, is discovered to refer to the place where the cake had been before it was
moved i.e. in the oven.
This type of temporary ambiguity should be more difficult than the others to
process. The main reason for this is the relatively low frequency of sequential prepositional
phrases in the English language. The reader simply has not come in contact with a sentence
of this structure before and so, does not expect it. This will be further discussed later in the
paper in connection with the results from the experiment in chapter seven.
4.3.5 Main Clause Vs Complement Clause
This category of ambiguity is relatively easy to explain. It should be more difficult
to process however. As has already been discussed in the previous sections, the main clause
of a sentence is where the main predication lies. It is only referred to in order to distinguish
it from other types of clauses that may be present in the sentence.
A complement clause is defined in section 4.2.1. In this instance of ambiguity, the
main clause is so termed to distinguish it from the complement clause in the sentence. With
both of these arguments sufficiently defined, it remains to show how they might be
confused when found together in a sentence.
18
Sentence taken from Sentence comprehension, www.ling.udel.edu
48
18.
The boat floated down the river sank without a trace19
Whilst the reader progresses through the sentence, the constraints of a main clause
are fulfilled by the part of the sentence at (19).
19.
The boat floated down the river
Like the other ambiguities however, when the next word is processed, a reanalysis
is required as it does not fit the initial interpretation. By the time the reader has reached the
end of the sentence, he realises that it was not in fact a main clause that had been read, but
rather a complement clause. The semantics of the main clause are therefore “the boat sank
without a trace”.
This type of temporary ambiguity might be expected to be the most difficult of them
all to process and gain a correct interpretation from. The main reason for this can by
hypothesised as the fact that this type of sentence construction is extremely rare in the
English language. A regularly preferred replacement for the example given would instead
be the unambiguous counterpart at (20).
20.
The boat that was floated down the river sank without a trace
This will be discussed again later when the experiment is the topic of the section.
4.4
ICECUP VERSION 3.0
Once the individual categories of temporary ambiguity have been decided upon,
further example sentences are required to satisfy the material needed for an experiment.
After all, more than just one example is needed of each category if we are to draw any
substantial conclusions from such an experiment.
In order to find enough material for an experiment, a corpus was used. In theory,
this allowed for a searching procedure to find different instances of the five categories of
temporary ambiguity.
The piece of software that was used to gain material for the experiment is called the
International Corpus of English Corpus Utility Program or in short ICECUP version 3.0.
The two previous versions of the software were developed by Akiva Quinn (1992) and
Porter & Quinn (1996) respectively. The third version was realised by Sean Wallis (1998)
which was designed from the outset for exploring the parsed version of ICE-GB, the related
corpus.
The software was received from Dr. Martin Emms and was gratefully employed to
find ambiguous sentences. From a CD ROM format, the program runs quickly and with
relative ease.
19
Sentence taken from Sentence comprehension, www.ling.udel.edu
49
ICECUP is a program that can achieve many different goals. For the purpose of
finding material for an experiment, it can be employed in a number of different ways. The
first is that it contains, as stated above, a parsed version of the corpus ICE-GB. From this
parsed version, a particular syntactic construction can be unified with a self constructed
syntax tree. This allows us to find all the similar syntactic constructions in the corpus. In
theory, relating that to the purpose of finding material for the experiment, we can enter a
syntactic construction of an ambiguous category and are able to find other examples that
can be used in the experiment.
To define such constructions in ICECUP is relatively easy to do. The button entitled
“New FTF” starts a new interface that graphically represents the syntactic tree you are
about to look for. Figure 4.4.1 illustrates how a syntactic tree might look after a few
functions and sentence constituents have been added. The head node is signified by the
yellow node to the top left. The “PU” stands for parsing unit and it stands as the root of any
tree you may require. The “CL” stands for the fact that you are searching for a clause.
Instead of the tree structure from top to bottom as is commonly used in linguistics, ICECUP
employs a left to right equivalent.
Figure 4.4.1
The five black arrows in between the nodes in the illustration signify that the next
node is directly following the other i.e. there cannot be any other node in between.
50
Figure 4.4.2
Once a syntactic tree has been decided on, the search button in the top right corner
of figure 4.4.1, entitled “start”, begins the exploration of the corpus. The results are then
displayed in a graphically pleasing manner as you can see from figure 4.4.2. The relevant
sentences can then easily be read off and added to the experiment.
The second way in which the software can be used to find material for the
experiment is the fact that it can search the corpus for not only a syntactic structure but also
a word. In order to search the corpus for a word, the big button entitled “text” at the top of
the page in figure 4.4.1 must be pressed. A small window appears with a text box in the
centre. It is here that a word may be entered. To begin the search, simply press the start
button.
Figure 4.4.2 also demonstrates that you can stop the search mid-way through
finding all the results. This can be achieved by pressing the start button whilst the program
is running a search. This can be quite an advantage if the set of results found is extremely
large and of which, only a small subset may be required. By pressing the same button, the
search can be reconvened.
In addition to this, ICECUP allows for the search of a word as part of a syntactic
structure itself. This makes the procedure easy to follow when searching for relevant
material. Below, in figure 4.4.3, a typical snapshot shows how to incorporate the search for
a word into the syntax tree.
Figure 4.4.3
51
4.4.1
Disadvantages
There are a number of disadvantages with the software however. The first is
probably the most important with respect to the intended experiment. Despite numerous
ways of being able to search through the corpus, not one ambiguous sentence could be
found. This, it must be assumed, reflects the unambiguous nature of language in general.
Ambiguity is generally avoided, unless intended, and this phenomenon results in the
unambiguous corpus.
In order to overcome this disadvantage however, the search can be concentrated on
the sentence’s unambiguous counterparts. For example, the syntactic structure that could be
searched for was a sentence like:
20.
The cook placed the cake that was in the oven on the table
In lieu of an ambiguous sentence such as:
21.
The cook placed the cake in the oven on the table
This type of approach achieves moderate success, enough to make it a viable means
of gathering material. As such, it is employed for the current experiment.
52
A further disadvantage with the software is that it does not use universally agreed
annotation in its syntax tree. For example, verbal constituents were not necessarily
dominated by a verb phrase as in figure 4.4.1.1 below.
Figure 4.4.1.1
This disadvantage can be overcome by some straightforward learning. For each
syntactic structure that is required, a structure with all the necessary components must be
found in the corpus. For example, if a structure is required where a noun phrase follows a
relative pronoun, then a similar sentence must be found in the corpus. This could be
achieved by searching for specific words in a sentences that corresponds to a noun phrase
followed by a relative pronoun such as “the man who”. By double clicking on the
identification number to the left of each of the resulting sentences, illustrated in figure
4.4.2, the syntax tree for that result sentence is displayed. From this example construction,
the syntactic annotation can be learned. This is a little time consuming but after the specific
syntactic quirks are realised, this disadvantage diminishes considerably.
Overall, ICECUP is a very useful tool to possess when dealing with language. It
can perform many tasks which range from looking for simple words to finding complex
syntactic structures. The program can be used to find unambiguous sentences. The
ambiguous sentences that are required can then be formulated from their unambiguous
53
counterparts. This is the method that was employed in order to find material for the
experiment.
4.5
Design of the Experiment
Taking all the previously described factors, recommendations and parameters into
account, an experiment has been constructed to test the temporary ambiguity resolution
potential of humans. More specifically, which type of ambiguity proves most difficult to
understand and process. A comparison of the results from these ambiguities can then be
used to highlight the differences between certain sections of the population that possess
varying level of resolution competency.
The experiment tests the relative time it takes for an ambiguous sentence to be
resolved. It is relative to the closest unambiguous copy of the sentence. For instance, if a
subject sees an unambiguous sentence such as:
22.
The milkman remembered the address that was left at home
Then he would also see its ambiguous counterpart, as in sentence (23). This relation
is reciprocal. Therefore if he was first to see sentence (23) then he would later see sentence
(22).
23.
The milkman remembered the address left at home
As the subject is reading through, say sentence (22), he will only see the sentence
one word at a time. This type of experimentation is used by a number of previous linguists
such as Traxler (2002). The subject needs to prompt the next word through a stimulation of
some sort. This is described in the next chapter under the topic of experimentation but for
now it suffices to know that a computer is employed for such a task. To stimulate the next
word, the subject must click a button with a mouse. After each click, the word that they
have read disappears whilst being replaced by the next sequential word in the sentence. To
illustrate this, the subject would progress through the sentence as described below:
The………………………………………………………………….
……milkman……………………………………………………….
………………remembered………………………………………...
……………………………...the…………………………………...
…………………………………..address………………………….
[Click of button]
[Click of button]
[Click of button]
[Click of button]
[Click of button]
54
……………………………………………that…………………….
…………………………………………………was……………….
………………………………………………………left………….
……………………………………………………………at………
……………………………………………………………….home.
[Click of button]
[Click of button]
[Click of button]
[Click of button]
[Click of button]
The same process is enacted for sentence (23), that is, when the time comes for the
subject to read it. Let us now introduce the term, “ambi-pair”, to cover this couple of
ambiguous and unambiguous sentences that a subject sees. For the above example, the
ambi-pair would consist of sentences (22) and (23).
In addition to being shown a set of ambi-pairs, each subject will finish the
experiment by being shown a paragraph in the same way as the ambi-pairs, one word at a
time. The paragraph will contain a few short sentences but will always have a temporarily
ambiguous sentence as its last. This ambiguous sentence will have appeared in an ambi-pair
of another subject. For example, if one subject had the ambi-pair of (22) and (23), then a
different subject may have the ambiguous sentence of (23) as the final sentence to their
paragraph.
This part of the experiment is to test if prior contextual knowledge might aid a
subject in the relative time it takes him to resolve the ambiguity. In other words, the relative
time it takes for a subject to resolve the final sentence in their paragraph can be compared to
the time it took another subject to resolve the exact same sentence but in an ambi-pair,
which is, the same sentence but in isolation. To revert back to a hypothesis of the parallel
model of section 3.5.2, if this was a more realistic model of the way in which we process
sentences, then you would expect the ambiguous sentence in the paragraph to be resolved
relatively more quickly than the one in isolation. To further specify how the experiment is
set up, it is first necessary to mention the five different versions construction.
4.5.1 Five Versions
The experiment is set up on the basis of five different versions. Each version
contains a different set of sentences. A participant in the experiment can only complete one
of the five versions. All five versions however share the exact same construction. If we
were to allow one subject per version then we would trivially have five subjects. Let’s call
them Subject (A)….Subject (E).
Subject (A) is shown five ambi-pair’s in total. Each one corresponds to a different
category of temporary ambiguity as described in section 4.3. In addition to this, he is shown
a paragraph. He makes his way through the paragraph in the same way as the sentences.
His final sentence however will be ambiguous. The same sentence is seen in an ambi-pair
of say, Subject (B).
Each subject sees only one paragraph as there is only one paragraph at the end of
each version. Each of the five paragraphs contains a different example of one of the
55
categories of ambiguity as described in section 4.3. For instance, Subject (C) may have a
main clause versus complement clause temporarily ambiguous sentence in an ambi-pair that
Subject (D) has as the last sentence in his context paragraph. To clarify what each of the
five subjects are exposed to, a general plan for the five versions is illustrated below. Let the
five categories of temporary ambiguity be Category (1) …Category (5):
Figure 4.5.1.1
AMBI-PAIRS
PARAGRAPH
Subject(A)
Category(1) Category(2) Category(3) Category(4) Category(5)
CATEGORY(5)
Subject(B)
Category(1) Category(3) Category(4) Category(5) Category(2)
CATEGORY(1)
Subject(C)
Category(3) Category(4) Category(5) Category(1) Category(2)
CATEGORY(2)
Subject(D)
Category(4) Category(5) Category(1) Category(2) Category(3)
CATEGORY(3)
Subject(E)
Category(5) Category(1) Category(2) Category(3) Category(4)
CATEGORY(4)
The above diagram shows the actual layout of the five versions. Subject (A)
therefore starts with a sentence from an ambi-pair of the first category. This is followed by
a sentence from the second category and so on. Subject (C) starts with a sentence from
category three, etc… In relation to the paragraphs, Subject (E) is shown an ambiguous
sentence from category four. The structure of the ambi-pair layout in each of the versions is
treated in section 4.5.3. The next section discusses the instructions that each participant
receives.
4.5.2 The Instructions
The instructions that each of the participants sees at the beginning of their version
of the experiment is the exact same, after all, the five versions differ only in the set of
sentences employed. As discussed in section 4.2.2, the instructions must be very specific
and clear. As described in the appendix, they are extremely short as well so as to not bore
the subject.
The instructions do not make any attempt to inform the participant as to the
motivation of the experiment in contrast to Greenbaum (1977b) suggestion. This is due to
the fact that syntactic and semantic processes are largely unconscious and any overt attempt
to inform the subject of the motivation behind the experiment would jeopardise the validity
of the results. This desire to keep the motivations behind the experiment hidden from the
56
participant is an ultimate goal throughout. The subjects are informed as little as is required
for them to complete the experiment successfully.
They are first informed of what they are about to see i.e. ten sentences and a
paragraph. They are then informed as to how they will see these sentences. An important
instruction given is the fact that all the sentences are fully grammatical. This will encourage
the user to treat the sentences at face value i.e. not to look deeper into the sentences than a
short glance allows. They are then informed of how they can progress through the
experiment. The instructions then state that questions will appear after the sentences. With
all this taken on board, the subject is equipped to begin the experiment.
4.5.3 The Sentences
The sentences that each subject sees are described in the appendix of the paper.
There is one aspect to the sentences however that has not been discussed so far. Which
sentence of the ambi-pair should the subject be exposed to and at what stage in their
version?
To get a second opinion on the matter, Dr. Martin Emms was consulted. The
outcome of the consultation was that each subject should start their version of the
experiment with an unambiguous sentence. This would introduce them to the make up of
the experiment as the sentence should not pose any difficulties to parse. So, for Subject (A),
he sees an unambiguous sentence from category (1) first. The ambiguous-unambiguous
sentences are then alternated so that Subject (A)’s second sentence is an ambiguous
sentence from category (2). His third sentence is then an unambiguous sentence from
category (3), etc…
To recap, the order as described in figure 4.5.1.1 is repeated for each subject before
a paragraph is seen. There are five ambiguous and five unambiguous sentences that follow
the order with an unambiguous sentence beginning the experiment. Each subject merely
sees a different example sentence of the temporary ambiguity categories.
In addition to reading through the sentences, the participants must demonstrate, at
the end of each sentence, that they resolved the ambiguity successfully. Therefore, at the
end of each sentence in the ambi-pair, a control question must be answered. These are all
documented in the appendix. In essence, the control question requires that the subject has
correctly parsed the sentence; otherwise the answer will be incorrect.
4.5.4 The Participants
The majority of the participants that took part in the experiment were Trinity
College students. Many Trinity College lecturers also took part. Originally, the links to the
57
experiment were intended for outside participation, family and friends, but due to technical
difficulties, discussed in the next chapter, outside use was prevented. This unavoidable
problem left the experiment short in relation to the premise that the participants must be a
representative cross section of the entire population. However, great efforts were gone to in
order to make the participants as representative as possible.
Friends and fellow students were the main source of participants to satisfy the
category of non-linguists in the population. Fellow CSLL students were the main source of
results for the category to represent linguists in the population. In addition to this, many
friends and lecturers helped out in order to record results for the bilingual analysis category.
Non-linguists were the most difficult category of the population to find. More precisely,
older non-linguists were the most difficult to get to do the experiment. As access to the
system is limited, often even within the college network, time had to be spent recruiting
middle aged non-linguists.
When the initial response was insufficient however, a new tactic was employed to
generate subjects. In order to increase the size of the experiment, making the results a better
representation of the population, and in order to recruit more non-linguists, time was spent
in computer labs throughout college. A small slip of paper was distributed to allow the
receiver to access the experiment on-line. The idea behind this will be discussed further in
the next chapter. It suffices to mention that in total, 110 slips were given out with moderate
success of hits to the site. This was the main method employed for getting non-linguists to
complete the experiment.
All the participants that took part had to fill in a personal form at the end of the
experiment. The details of the form are left to the next chapter. For now it suffices to say
that the questions asked of the subjects are done so in order to categorise them during in the
consideration of the results. Their linguistic background was asked of them along with
whether they were male or female. Their age is also asked of them so that a representative
population can be simulated as much as possible i.e. to reduce the domination of studentaged participants.
Chapter five
Experimentation
58
5.1
Introduction
Experiments can be quite difficult to get right. As described in the previous
chapters, there are a lot of parameters that must be taken into account. The experiment that
is described in the previous chapter has a further requirement. This requirement however, is
the most important one.
In order for the experiment to be achieved successfully, a machine must be
employed to handle the experiment. The obvious machine that is used is a computer. A
computer must be used for the experiment for several reasons. Firstly, computers can be
used to not only run the experiment but to also store all the results of the experiment. They
can be easily manipulated to run efficiently and effectively. The second reason, and
probably the most important one, is in relation to the specifics of the experiment.
59
Time is the essence of the experiment, more specifically, time intervals. The
experiment is not primarily testing answers to questions as such or how someone thinks; it
is simply testing the time it takes to read sentences. A sentence may take any amount of
time to read, but words take a much shorter length of relative time. In fact, a word is
generally processed in a split second, depending on its length and complexity. It is virtually
impossible to time a person, word by word, reading a sentence. You would be starting and
stopping a stopwatch far too quickly to gain any information from the exercise. Therefore,
without the aid of a computer, this type of experiment would prove virtually impossible.
A computer is, on the other hand, well equipped for the purpose of timing. As the
participants progress through the sentence they see one word at a time. To see the next
word they must stimulate the computer into action through the pressing of a button. For a
computer, timing a person’s response to a button stimulus is simple. In this sense, a
computer can be employed to do nearly all of the experimentation work.
This chapter discusses the method of experimentation. A computer is used as the
basis for the experiment. To make the experiment more accessible however, the program
that runs the experiment is not reliant on software being run on a single computer. A webbased tool is employed so that the experiment can be accessible from a theoretically
enormous source, the entire world. As a result of difficulties faced, this was not possible for
the current experiment but in theory, any number of participants from all over the world
could participate.
5.2
Web Tool
To implement the experiment, a web tool was employed. The web tool that is
referred to throughout the remainder of the paper is an automatic web-based
experimentation tool that has been previously developed by past final year students of
Computer Science, Linguistics and a language. Six students have previously worked on or
with the tool as part of their Final Year Project (see Kenny(1998), McGowan(1999),
Guennouni(2000), Ryan(2001), Hourihane(2002) and Harrow(2003)). Sarah Kenny
originally worked on the idea in 1998. The motivation for the project in the first place was
Kenny’s interest in working on a multidisciplinary project that encompassed intense
programming in a psycholinguistic domain. In this, she maintained a balance between
computer science and linguistics, a concept on which the degree is based. The idea for
creating such a web based experimentation tool is down to the fact that:
“no serious tool exists for those involved in creating linguistic experiments to model their
experimental work on the web”20
In more recent years, Fionnuala Hourihane (Hourihane 2002) improved upon the
initial design. She added the function of allowing multiple experimenters, a facility that is
very important for the accessibility of such a program. She also designed an Experiment
20
Quote taken from A Generic Automatic Experiment Creation and Presentation Tool, Kenny (1998)
60
Manager to oversee these experimenters and the entire system. Yet another facility that she
added was a system that provided user authentication to control the access to the
experiment, this is very important so as to control the participants that take part in an
experiment and in doing so, to safe guard the authenticity of the results. She achieved other
successes as well such as creating a list of previously created experiments along with their
details, a redesigning of the file storage system, an editing of experiments facility and a
facility to allow the experimenters to view and check their own experiments without
recording any results for themselves. In all, Fionnuala refined the web tool so that it could
be used more easily and effectively by possible experimenters.
Gratitude must also be extended to Yvette Graham who is currently under going a
Masters by research program with the computational linguistics group in Trinity College
Dublin. Yvette acted as the Experiment Manager for the current experiment. She is
currently working on the web experimentation tool by further adding the facilities to
analyse experiments, edit them and to greater improve the user interface.
The amalgamation of all the stated ideas and subsequent improvements has resulted
in a web tool that has enormous potential and functionality. There are several advantages
for using such a tool. Some of these have already been alluded to but they will now be
explained in more detail.
Hoy&Sheehan (1999) outline seven advantages of using the internet to conduct
experiments. The first of these is design flexibility. Design flexibility refers to the fact that
experimenters:
“can take advantage of the graphic power available through programming languages”
Another advantage, referred to in the previous section, is that it allows a much
greater audience to partake in the experiment. This advantage is coupled with the possibility
of comparing dialectal variations of the language experiment. The web based tool also
encourages the participant to partake in the experiment at their leisure i.e. in their own time
and space. This generally makes the results more accurate and in doing so improves the
conclusions that can be drawn from the experiment.
The internet also cuts down on the costs to the experimenter in both time and
resources. A high number of responses can be generated relatively quickly. In addition, the
experimenter is not required to be present during an on-line experiment which allows him
more time to generate possible subjects and ultimately analyse the results.
There are many advantages connected to the absence of the experimenter from his
experiment. Subject anonymity is the most important of these. The subject is unknown to
the experimenter which improves accuracy of results even further and can also cleanse the
experiment of any biases that the experimenter may have with respect to the individual
participants that he would otherwise have had to meet, face to face. With all the inherent
advantages, the current experiment was made available to be conducted on any computer
that could gain access to the Internet. In practice however, due to technical difficulties it
was largely confined to the Trinity College network. The specifics of the web tool and how
it was used to achieve this are described in the following section.
61
5.3 Using the Tool to Define the Experiment
To begin the process of constructing the experiment, contact was made with Yvette
Graham, the Experiment Manager. A link was attached to her responding e-mail whereby
the program could be accessed. A login name and password was also attached. This
information was then used to gain access to the login in page and subsequently to the home
page.
The first page that is displayed is the personalised home page for the experimenter.
It contains all the experiments that the individual experimenter may be using. There is no
limit on the number of experiments as they are all stored on the computational linguistics
research groups file space, but the names of the experiments must have no spaces contained
in them. Armed with this information, the construction of the experiment can begin. In
figure 5.3.1, the final version of the homepage can be seen. There are a couple of practice
experiments entitled “TempSubject1 ”, “Temp” and “Tempguity ”, etc… which were not
used but were vital to the learning process involved in using the tool.
Figure 5.3.1
62
Experimenter’s homepage. Subject one, two, three, four and five represent the five versions of the experiment.
Some of the others represent the trial and error files.
The five versions of the experiment as described in section 4.4.1 can be seen in figure
5.3.1. They are named “Subject1”-“Subject5”.These will be discussed again later in this
chapter.
The tool caters for a variety of different styles of experiments. Each experimenter
that uses the system may require an Experiment Manager to taper the program to the
necessities of the individual experiment. After a brief self introduction to the program, a
meeting was arranged with Yvette. The outcome of the meeting was to clarify the needs of
the experimenter for the experiment and also what the limits of the web tool are. The
accessibility was explained and the basic platform of the experiment was decided.
5.3.1
The Ambi-pair Design
63
The tool is easiest described as a basic platform of sequential slides. On each slide,
there are several options available to the experimenter. The specifics of the current
experiment require that a sentence be shown one word at a time. Every word was therefore
represented by an its own individual slide. The sentences were modelled around a series of
slides which together make up a section. Each section corresponds to a different sentence as
shown in figure 5.3.1.1.
Figure 5.3.1.1
SECTION ONE
Slide1 Slide2 Slide3
Slide4 Slide5 Slide6 Slide7 Slide8 Slide9
The
over
man
fell
the
tall
lady’s plastic
bag
The options available to the experimenter on each slide are illustrated in figure
5.3.1.2. These are the options after a passage for the slide has been selected and the word
“the” has been typed to represent the slide. The option to add a passage was chosen each
time to represent a word. The passage in each of the sequential slides therefore contained
the words to the sentences as shown in the example of figure 5.3.1.1 above. Initially, the
tool was constructed in a way that required each slide to contain a question. The
Experiment Manager soon alleviated this draw back.
On the last slide, an additional option was used however. This was to facilitate the
presence of an ambiguity control question. The final slide in each section contained a
question. The questions are only strictly required after the ambiguous sentences to show
that the sentence had been adequately understood but through trial and error the experiment
was constructed with a question at the end of every sentence. This was done so that the
subject was given a break after progressing through each sentence.
The first version of the experiment just contained questions at the end of the
ambiguous sentences. This was decided to be inefficient as the participants began the next
ambiguous sentence straight after the unambiguous one which would have belittled the
unambiguous sentence in the eyes of the subject and may have ultimately distorted the
results. After all, the times of both sentences in the ambi-pair are equally important as they
are compared to one another in the results.
Figure 5.3.1.2
64
Slide page. “The” has been entered to represent this particular slide through pressing the “add passage” button.
As described in the previous chapter, the experiment was constructed to contain five
versions. These five versions represent five different subjects as defined by the web tool
and illustrated in figure 5.3.1. Each subject sees a different set of experiment sentences.
Each of these sentences must be added in the manner described by figure 5.3.1.1. Each of
the participants sees five ambi-pair categories which equates to ten sentences. This
corresponds to ten sections of a varying number of slides, depending on the number of
words in each sentence. There is however an eleventh section. This corresponds to the
context paragraph as described in the previous chapter.
5.3.2
The Context Paragraph Design
The context paragraph is set up in similar fashion to the ambi-pair sentences. Each
slide corresponds to a single word in the paragraph. The difference with this final section is
that there are many sentences before an ambiguity control question is asked. The question is
still asked in the last slide of the section but there have been many sentences that have
already been read by the subject without any question at the end of them. These previously
read sentences should construct a gist as explained in section 3.4, which could aid the
resolution of the final ambiguous sentence. This will be further discussed again in section
65
7.7. A sample slide with a question is illustrated in figure 5.3.2.1 below. The same layout of
the page is relevant to any of the previously mentioned slides that contain a question such
as at the end of each ambi-pair sentence.
Figure 5.3.2.1
Question page. Sample ambiguity control question and answer.
With the treatment of both ambi-pairs and the context paragraphs, the experiment is
nearly ready to be run. It remains to explain the construction of the form that is used to gain
personal details form the participant. This information is the key to the possible explanation
of results and the conclusions that can thereby be drawn.
5.3.3
The Form Design
66
The form is an essential part of the experiment. It gathers participant data that will
ultimately be used in the analysis of the results in chapter seven. A form can be placed at
the beginning or end of an experiment. For the purpose of the current experiment, the
personal form was put at the end. There was an important reason for constructing it like
this. Temporary ambiguity resolution is a mainly subconscious event. Therefore, the less
they know about the experiment before completing it, the more accurate the results will be.
If the subject is aware of the aims of the experiment on the other hand, some unexpected
conscious decisions may obstruct their results. On this principle, the form was put at the
end of the experiment. If the participant filled the form in beforehand, they may be biased
or prejudice in some way towards the experiment knowing that their details are being used.
Figure 5.3.3.1
Form page. Personal questions that each participant must fill in at the end of the experiment.
The form is constructed to gather all the relevant information from each participant.
Data for the areas of comparison must be acquired. This is achieved by asking questions as
to their linguistic background to see if they can be classed as a linguist. The fact that they
are male or female and whether they are bilingual or not will also be compared in the
results. With the different classes of participants, one further control question is required.
To try and keep the set of subjects as close to an accurate cross section of the population as
possible, the age of the subjects is asked. To keep a good representation, there must be an
67
equal number of different aged participants; the age field in the form relays this information
to the experimenter. The graphical layout of the form is illustrated in figure 5.3.3.1.
As one can see from the illustration, there is a fifth and final question. This was
included to see if the participant was aware of the aims of the experiment. This is crucial to
know as the unconscious nature of the experiment is vital. Linguists, by definition, should
naturally be more conscious of the unconscious and therefore one would expect a greater
accuracy of guessing for this set of subjects. The final question is therefore a control
question to insure that the results are as accurate as possible.
5.4
Participation
As discussed in the previous chapter, the participants varied in their backgrounds.
Participation in the experiment is quite straight forward however and this is an advantage in
relation to older subjects who may not be too familiar with a computer as means of
experimentation.
To participate in the experiment, three items were needed; a web site address, a
login name and a password. At first, these three items were supplied to possible subjects via
e-mail. Due to a poor hit rate however, a new approach can be taken. In figure 5.4.1, an
example slip of paper can be seen. It contains the three items that are required to participate
in the experiment.
Figure 5.4.1
wilde.cs.tcd.ie/Experimenter/Version1/Data/canice/Subject1/participate.html
Login name: experimentmethere
Password : nice1
After the subject has completed the experiment, a result file is automatically created that
can be accessed from the experimenters home page. The participant does not have any more
access to the system after they have finished. There is however, one participant that can do
the experiment without recording results i.e. the experimenter. From his home page he can
decide to run the experiment to check it before any participation. This procedure was
enabled by Hourihane (2002) and is an extremely useful addition.
68
A final aspect to the participation that needs to be mentioned is that there are five versions
of the experiment as described in section 4.5.1. These five versions are saved as different
files on the system. A participant is simply given an address that contains access to just one
of these five files decided by the experimenter. Each different file requires a different login
name and password. For simplicity, all five login names were kept the same, namely
“experimentmethere”. The passwords however were numbered from nice1 to nice5
depending on which number file was being accessed. So, for example, if a person was
intended to complete version 4 of the experiment, they were distributed with a piece of
paper as in figure 5.4.1 but with slight numerical difference. In their url address, “Subject4”
would appear instead of “Subject1” and the password would read “nice4” instead of
“nice1”.
With all this in place, the experiment can be conducted. Any participant that accesses the
system will have their details saved on file. At the end of the experiment, the results can be
calculated. To achieve this, another meeting was necessary with the Experiment Manager to
taper the system to the experimenters needs.
5.5
Viewing the Results
After a version of the experiment has been completed for the first time, a new
results directory is created and their details are added to it in a new file. The next time that
version of the experiment is completed, a new result file for that subject is added to the
existing directory. In order to view these results, there is an option on the experimenter’s
homepage to view the results of a particular experiment. This page leads to another one
which queries whether the results are to be displayed by subject or by question. If you
choose to view them by subject, the resultant page is illustrated in figure 5.5.1. Here you
can specify which subject’s results you wish to view.
As the system originally stood, you were either able to view the resultant slides of
the experiment that the participant had filled in, or view all the answers to the questions for
that experiment. So, in relation to the current experiment, the experimenter could only view
the answers to the ambiguity control questions and the form at the very end. There was
however, no facility for the experimenter to check the type of results that the current
experiment requires i.e. a temporal analysis of every slide contained in the version of the
experiment. A brief meeting with the Experiment Manager sufficed to get the results that
are required for the current experiment. The system already contained a facility that times
slide exposure but this was not initially displayed in the results. Yvette Graham was able to
alter this so that a temporal analysis file was enabled.
Figure 5.5.1
69
Result directory page. The five subjects that completed version one of the experiment can be seen in the
diagram. By clicking on one of them and pressing the “continue” button, the ambiguity control questions and
the form for that subject can be viewed.
5.5.1
The Temporal Analysis
The addition to the system comes in the form of the temporal analysis of the slides.
This was achieved by altering the program to contain an option to view the times associated
with each slide that a particular participant saw. The system already contained options to
view the results by subject or by question. The option to view the times was simply added
to this page.
To view the individual times however, a little extra bit of work is required. The
temporal results come in the form of a “.csv” file. These files contain all the times
corresponding to the exposure that a particular subject received to each slide. They come in
a particular format which separates the data by a series of commas and tabs. These files,
once copied and pasted to notepad, could then be loaded into a Microsoft Excel platform.
The files were designed by the Experiment Manager to load into Excel in such a way as to
assign each value into the correct row and column. Figure 5.5.1.1 shows a typical “.csv”
file saved as a notepad text file.
70
Figure 5.5.1.1
The Excel program allowed the viewing of the temporal analysis in an eloquent
way. Excel functions could then be easily run on the figures to demonstrate graphs and bar
charts that will be discussed in the next two chapters. Figure 5.5.1.2 shows the “.csv” file in
an Excel spreadsheet.
Figure 5.5.1.2
71
The only change that was made to the files in figure 5.5.1.1 and 5.5.1.2 was to edit the
column names. As the files show, the column names were assigned the section and slide
identification of the experimentation tool. In order to improve the quality of the resultant
graphs, these names were systematically replaced by the actual words that they represented
in the experiment. In achieving this, the graphs could be read more easily. The portion of
the text file that contained the identification names were simply deleted and replaced, each
word being separated from the next by a comma. The file could then be loaded into an
Excel platform in the same manner as before.
5.6
Problems Encountered
There were numerous problems encountered with the web tool. As mentioned in the
previous chapter, technical difficulties were a large obstruction to the experiment running
smoothly. The tool is simply not robust enough and needs to be treated very delicately. It
often tended to crash at the slightest mistake such as giving two experiments the same name
or pressing the “Back” button on the web browser. In the case of it crashing, the system had
to be restarted from the server which caused long delays and annoyance in setting up the
experiment.
72
The tendency to crash during the setting up of the experiment could be overcome by
persistence and patience but one of the biggest drawbacks to the experimentation tool as a
whole is that participants do not have the same patience as the experimenter. After all, they
gain very little from completing the experiment. As a result, the hit rate to the experiment
was extremely poor. The participants complained of many difficulties faced. One such
problem was simply being faced with a blank screen, indicating that the system had
crashed. Another complaint was that only a grey box could be viewed, this indicated that
the computer would not run the program for an unknown reason. In other words, the applet
that should have opened in order to view the experiment simply didn’t load. Yet more
complaints were that the web page didn’t load at all, this can be explained by certain
browsers that are incapable of viewing the page from outside the college network.
The patience required to use the system cannot be asked or expected of the participants.
Due to this, many possible subjects were lost and this in turn affects the accuracy of the
results as you have a much smaller set of subjects with which to represent the entire
population. There were however, enough successful participants to deem the experiment a
success and the persistence on their behalf is greatly appreciated.
A final aspect of the tool that proved troublesome was during the construction of the
experiment. Quite simply, there was no means of editing an experiment that is under
construction. For example, five ambi-pairs and several sentences in the context paragraph
needed to be entered for each version. Each sentence was made up of a group of slides as
described in figure 5.3.1.1. If any of the slides were constructed incorrectly or a word spelt
incorrectly and the button had been pressed to begin the next slide, it was too late to correct
it. In other words, every word of every sentence had to be entered correctly with the correct
questions on the right slides or otherwise the procedure has to be started all over again,
wiping what had been constructed to that point. If this happens at the end of the context
paragraph for instance, the maximum time is lost. This proved to be quite time consuming.
5.7
Possible Improvements
Despite all the disadvantages to the system as laid out by the previous section, the
web tool proved to be extremely easy to use. It is very user friendly and no previous formal
computing instruction is necessary in order to use the program. In this sense, it is the ideal
tool to use in order to create an experiment. Without it, an experiment of this nature would
have been virtually impossible to achieve.
The most obvious and essential improvement to the system would be to facilitate the
editing of experiments. The time it takes to construct an experiment is needlessly long. An
option to navigate through the experiments, with the possibility of changing parts where
required, would be a huge advantage if it was added to the system.
Another advantage would be to link the experimenter’s homepage with any subpage of the homepage. For instance, after an experimenter has finished an experiment he
73
needs to retype the address to get back to the homepage. A little shortcut button would
make a big difference.
Improvements to the overall robustness of the system are very important for the
future uses of the web tool. Its tendency to crash over the slightest of irregularities needs to
be attended to. The web tool is still in its prototype phase and further extensions must be
added to it in order to make it the all round experimentation tool that it is intended for. It is
however, more than adequate for the purpose of the current experiment and gratitude must
be extended to all those who have been involved in its creation.
Chapter 6
74
Results
6.1 Introduction
The Web based experimentation tool records all the results that are needed to analyse
the way in which the sentences are processed. Each of the five versions of the experiment
received a varied number of subject participation. Below in figure 6.1.1 is a chart that
illustrates the amount of subjects a particular version received.
Figure 6.1.1
75
Frequency of participation
Series1
10
8
People
6
10
4
6
7
7
8
2
0
Version 1 Version 2 Version 3 Version 4 Version 5
Experiment
As the chart illustrates, 38 participants took part in the experiment. The distribution of
participation in each of the different versions of the experiment was satisfactory. According
to the experiment parameters as outlined in section 4.3.1, Schütze suggests that in order to
validate conclusions, a diverse cross section of the entire population must participate.
Figure 6.1.2 illustrates the age distribution of the participants.
Figure 6.1.2
Participant Ages
Series1
30
25
20
People 15
10
5
0
16- 21- 26- 31- 36- 41- 46- 51- 56- 6120 25 30 35 40 45 50 55 60 65
Ages (years)
Unfortunately, this chart illustrates that the distribution of age was not satisfactory
and therefore cannot represent a valid cross section of the population. This occurred for a
number of reasons. The first of these is that many of the participants were students. As the
experimentation system was not robust enough to handle off-campus participation, the
natural consequence to the experiment was that only participants of student age took part.
On the other hand, figure 6.1.3 below, demonstrates that the distribution between the sexes
was satisfactory for the experiment.
Figure 6.1.3
76
Participant Sex
Series1
30
25
20
People 15
10
5
0
20
18
Male
Female
Gender
Both of these charts could be plotted with data gathered from the form filled in by the
subject at the end of the experiment. Two other areas were of significance to the experiment
at hand. To compare data between linguists and non-linguists, a portion of the participants
must have studied linguistics before. Figure 6.1.4 illustrates that the frequency of linguists
in the experiment was satisfactory.
Figure 6.1.4
Linguistic orientation
Series1
25
20
People
15
10
24
14
5
0
Linguists
Non-Linguists
Another factor that has to be taken into account is whether or not the particular
subject was a native speaker of English, the language of experimentation, or not. Once
again, data from the form could be plotted resulting in figure 6.1.5 below.
Figure 6.1.5
77
Bilinguals
Native Speakers
Bilinguals
17
20
15
14
People 10
4
5
0
3
Male
Female
Gender
Figures 6.1.4 and 6.1.5 both illustrate that there is enough data from both these areas to
draw conclusions from their results. Just over one third of the subjects had studied
linguistics before. A preferable portion might have been half and half but this proved too
difficult to attain. Likewise in relation to bilinguals, just under a third of the males and just
over a sixth of the females were bilingual. The female bilingual figures were a little
disappointing compared to the males but overall, enough data is present to analyse the
results under these categories. The poor age frequency is the only area that is cause for
concern.
6.2
Structure of Results
The results to the experiment are dealt with in respect to two areas. They are
separated so as to juxtapose the relevant data for future comparison. The first part of the
experiment is concerned with the ambi-pair, more specifically, the processing difference
between the pair. There are five versions of the experiment and so there are five example
ambi-pairs per ambiguity category that can be studied. The following five sections of this
paper correlate to the treatment of these ambi-pairs one category at a time.
The second part of the experiment is concerned with the ambiguous sentences that are
found at the end of the context paragraphs. The data from the isolated ambiguous sentences
of the ambi-pairs is then treated with respect to the data gathered from the context
ambiguous sentences. The results for both parts are discussed in relation to two major data
sources. These are:
A.
The answers to the ambiguity control questions and
B.
The temporal analysis of each sentence in the experiment
78
The ambiguity control questions are designed so that the participants can demonstrate
whether they grasped the true meaning of the sentence or not. An important implication that
can be applied through the analysis of these results is whether the failure to satisfactorily
gain the correct syntactic interpretation is reflected in the way they process the sentences.
This is dealt with further in the next chapter.
The results to the temporal analysis are also an important source of data. They signify
the way in which the different sentences were processed by each subject. Difficulties
encountered by a particular participant can be seen from these results. This is important as,
to resolve temporary ambiguity, one would expect a slight temporal discrepancy at a certain
stage in the ambiguous sentences when compared to its unambiguous counterpart. The
graphs of all the sentences can be found in the appendix, category by category. The results
to the temporal analysis can be further split into two other data sources. These are referred
to when the results to each category is discussed. The two sources of data are:
B1.
The pauses by each subject in selected areas where ambiguity resolution is
expected and
B2.
The time it takes for the subjects to respond to the question at the end of each
sentence.
The pauses at certain points in the sentences are recorded as the percentage of
subjects that paused, ever so slightly (+0.5sec), after reading the word in question. The
selected part of the sentence is where the resolution of the ambiguity is expected. In order
to supplement this data, the second part of the temporal analysis may prove equally
significant.
The final word of every sentence was constructed to be on the same slide as the
ambiguity control question. The time of exposure to this slide is important as it shows how
long it took the subject to answer the question. The significance of this time is that even
though the temporal analysis of the sentence may not show any significant results from
word to word, it may show it in the time it takes them to answer the question. A reason for
this might be that the subject just whisks his way through the sentences and resolves the
ambiguity at the end, having resorted to his working memory, see section 3.4, to memorize
the words.
Throughout the next five sections, a number of tables are used to record the
correctness of the ambiguity control questions. In every case, a 1 signifies that the subject
gave a correct answer and a 0 signifies an incorrect answer. The two columns for each
“version” in the table represent the answers to both ambi-pair questions. The sub-column
on the left represents the unambiguous sentence, whereas the one on the right represents the
ambiguous sentence. The “C” signifies the answers to the context paragraph which only
contains an ambiguous sentence. This format will be kept throughout this chapter. In total,
the grid represents 76 responses to questions (38 participants * 2 ambi-pair questions). 38
of these are unambiguous sentences and 38 are ambiguous ones. In addition, the grid
contains a varying number of context responses as it depends on which version completed
the respective category as to how many response it gained.
79
The graphs referred to in this chapter that can be viewed in the appendix were constructed
using Microsoft Excel 2000. The text files discussed in the last chapter were simply loaded
into an Excel spread sheet and there is a built in function to produce charts from the data
stored in its rows and columns. It is an extremely powerful program that enjoys
multifunctional uses.
There are several points to note about these graphs. Firstly, the time it takes for the
instructions to be read by a subject is included. Each version began with a different
category which means that the instructions data will be present at the start of only five of
the sentences. This can aid in the analysis as it is clear which sentence the subjects see first,
a factor that may play a role in their processing of the sentence.
6.3 Complement Clause Vs Relative Clause
This type of temporary ambiguity is typified by the double use of the relative pronoun
“that”. The unambiguous sentence does not use this construction in favour of the passive
voice to represent the complement clause of the sentence. The reader should originally
miss-analyze the syntax of the sentence in favour of a complement clause and soon realize
this not to be the case. Figure 6.3.1 shows the results to the ambiguity control questions. In
this case, the context paragraph was seen by version two subjects.
Figure 6.3.1
80
Complement clause Vs Relative clause
Version 1 Version 2 Version 3 Version 4 Version 5 C
Subject 1
1
1
0
1
1
1
1
1
1
1
0
Subject 2
1
1
1
1
1
1
0
0
1
1
0
Subject 3
0
1
0
1
1
0
1
1
1
1
0
Subject 4
0
1
1
1
1
1
0
1
1
1
1
Subject 5
1
0
1
0
1
0
0
1
1
1
0
Subject 6
1
1
1
1
1
1
0
1
1
1
1
Subject 7
1
0
1
1
0
1
1
1
0
Subject 8
1
1
1
1
1
Subject 9
0
0
1
Subject 10
1
1
0
These results can be plotted in a bar chart as illustrated in figure 6.3.2 which shows the
trend in relation to how the subjects handled the correct resolution of the temporary
ambiguity they encountered.
Figure 6.3.2
Compl CVs Rel C
Ambiguity control questions
Unambiguous
40
30
28
Questions 20
Ambiguous
31
10
10
0
Correct
7
Incorrect
What is immediately striking about these results is that more subjects correctly answered
the questions relating to the ambiguous sentences than the unambiguous ones. This might
81
suggest that the sentences containing this type of temporary ambiguity are more easily
understood. Overall, there is not a large difference between the two which also implies that
those who were able to retrieve the correct information from the unambiguous sentence did
likewise in the ambiguous one. Roughly three quarters of the participants answered both
sets of control questions correctly which implies a satisfactory level of ambiguity
resolution.
Studying the results to the temporal analysis (graphs 1V1A – 1V5B), one can see a pattern
emerging. In this category the graph must be studied around where the relative clause
begins as this is where the syntactic reanalysis is expected. The diagram below shows the
percentage of participants who paused after reading the words denoted by “-----“.
Unambiguous
32%
3%
5%
16%
-----
that
-----
Ambiguous
32%
16%
----11%
24%
These figures show that after reading the start of the relative clause, a much larger
portion of participants required a brief pause in one of the subsequent words, this was a
third of the subjects on seeing the word “that” alone. The figures also show the percentage
of pauses in the unambiguous sentence before the word “that” is the same as the percentage
in the ambiguous sentence after processing the word “that”.
6.4 Object Noun Phrase Vs Main Clause Subject
This type of temporary ambiguity is typified by the starting of the sentence with a word
such as “while” or “when”. It is distinguished from the unambiguous sentence by not
including a comma mid way through the sentence. The reader is expected to miss-analyze
the ambiguous sentence by initially expecting the actual subject of the sentence to be the
object of the qualifying “while” or “when”. The control question data is as follows:
Figure 6.4.1
82
Object Noun Phrase Vs Main Clause Subject
Version 1 Version 2 Version 3 Version 4 Version 5 C
Subject 1
0
0
1
0
1
1
1
1
1
0
1
Subject 2
1
1
1
0
1
0
1
1
1
1
1
Subject 3
1
0
1
0
1
1
1
0
1
1
0
Subject 4
1
0
1
1
1
1
1
0
1
0
1
Subject 5
1
1
1
0
1
1
1
0
1
0
0
Subject 6
1
0
1
1
1
1
1
0
1
0
0
Subject 7
1
1
1
0
1
0
1
0
0
Subject 8
1
1
1
0
Subject 9
1
1
Subject 10
1
1
Once again, these results can be plotted into a chart which demonstrates the trend in
relation to the resolution satisfaction of the sentences.
Figure 6.4.2
Obj NP Vs Main C sub
Ambiguity control questions
Unambiguous
40
Ambiguous
37
30
Questions 20
10
0
21
17
1
Correct
Incorrect
83
As the figure 6.4.2 illustrates, nearly every unambiguous sentence question was answered
correctly. This figure was reduced by over 50% for the ambiguous counterparts. With more
subjects answering the ambiguous question incorrect than correct, it implies a poor level of
overall ambiguity resolution.
Looking at the resulting graphs of the temporal analysis (graphs 2V1A – 2V5B), we can
study the times around the words where the miss-analysis is expected. For this category, it
begins with the word before the main clause begins. The beginning of the main clause is
denoted by the comma present in the unambiguous sentence, but not the ambiguous one.
The triple bar represents the processing area after which a possible re-analysis would be
anticipated in the ambiguous sentence. The diagram below shows the percentage of
participants who paused over the relevant word.
Unambiguous
26%
13%
11%
11%
--------|||
-----
Ambiguous
11%
,
--------3%
3%
13%
24%
18%
These figures show a significant increase in pauses where expected, at the point where a
reanalysis of the syntactic structure of the sentence is required. Interestingly, the percentage
of pauses in the ambiguous sentence is very similar to the percentage produced by the
presence of a comma in the unambiguous sentence.
6.5 Direct Object Vs Embedded Subject
This type of temporary ambiguity is typified by a sentence that uses a verb that can be
syntactically analyzed in more than one way, such as “to suggest” or “to believe”. The noun
phrase immediately after the use of the verb is expected to be the object of this verb but in
reality it is the subject of the embedded sentence. This is distinguished from the
unambiguous sentence that employs the word “that” directly after the verb to allow the
reader to know that an object will not be immediately present. The results for the ambiguity
control questions are illustrated below.
84
Figure 6.5.1
Direct Object Vs Embedded Subject
Version 1 Version 2 Version 3 Version 4 Version 5 C
Subject 1
1
1
1
1
0
1
1
1
1
1
1
Subject 2
1
1
1
1
0
1
1
1
1
1
1
Subject 3
1
1
1
1
1
1
1
1
1
1
1
Subject 4
1
1
1
1
1
1
1
1
1
1
1
Subject 5
1
1
1
1
1
1
1
1
1
1
1
Subject 6
1
1
1
1
1
1
1
1
1
1
1
Subject 7
1
1
1
1
1
1
1
1
1
Subject 8
1
1
1
1
Subject 9
1
1
Subject 10
1
1
This data can be plotted in a similar manner to the previous categories to accommodate
future comparisons that will be discussed in the next chapter.
Figure 6.5.2
85
Dir Obj Vs Embedded sub
Ambiguity control questions
Unambiguous
40
36
Ambiguous
38
30
Questions 20
10
0
2
Correct
0
Incorrect
What stands out about the chart in figure 6.5.2 is that nearly everyone answered the
unambiguous questions correct. This is reiterated in the ambiguous sentences where not one
subject failed to resolve the temporary ambiguity. These results suggest that the resolution
of this type of ambiguity poses little problems to people, regardless of whether temporary
ambiguity is present.
Turning our attentions to the graphs corresponding to each of the sentences in the appendix
for this category (graphs 3V1A - 3V5B), a diagram can be constructed, once again to show
the percentage of pauses at certain points in the sentences. The study in both cases begins at
the word immediately before the noun phrase which constitutes the embedded subject of the
sentence. For the unambiguous sentences, this word is always “that”.
Unambiguous
0%
5%
8%
11%
3%
-------------
the
|||
-----
11%
16%
14%
11%
Ambiguous
5%
These percentages show a definite trend. The expected increase in pausing due to the
necessity of re-analyzing the syntax of the ambiguous sentences is denoted as coming after
86
the triple bar in the diagram. Interestingly, nothing was recorded to show that this was the
case. The percentage of pauses remained constant throughout the processing of this area of
the sentences. The only difference is that the words in the ambiguous sentences were
paused on much more regularly at the selected area.
6.6 Prepositional Phrase Attachment
This type of temporary ambiguity is represented by two prepositional phrases side by side
in a sentence. The ambiguity arises out of the unfamiliar nature of such a construction. The
difference present in the unambiguous sentence is the addition of the words “that was” to
denote the first prepositional phrase.
Figure 6.6.1
87
Prepositional phrase attachment
Version 1 Version 2 Version 3 Version 4 Version 5 C
Subject 1
1
1
1
0
1
1
1
1
1
0
0
Subject 2
1
1
1
1
1
1
1
0
1
1
0
Subject 3
1
1
1
0
1
0
1
1
1
1
0
Subject 4
1
1
1
1
1
1
1
1
1
1
0
Subject 5
0
0
0
1
1
1
1
0
1
0
0
Subject 6
1
1
1
1
1
1
1
0
1
1
1
Subject 7
1
1
1
1
1
0
1
1
0
Subject 8
1
1
1
1
0
Subject 9
1
0
Subject 10
1
1
Figure 6.6.2
Prep phrase attach
Ambiguity control questions
Unambiguous
40
Ambiguous
36
27
30
Questions 20
11
10
0
2
Correct
Incorrect
The quality of the control answers are represented by the chart in figure 6.6.2. Once
again, the questions to the unambiguous sentences proved to be of little difficulty as you
would expect. There is however, a 25% decrease in correct answers when considering the
88
ambiguous sentences. This suggests that a quarter of the subjects found difficulty in
resolving this type of ambiguity.
The temporal analysis of the sentences (graphs 4V1A – 4V5B), illustrated by the relevant
graphs in the appendix, can be studied in the area surrounding the expected reanalysis. For
this type of ambiguity, the selected area begins with the word before the start of the second
prepositional phrase. The percentages were as follows:
Unambiguous
16%
8%
3%
----preposition
----Ambiguous
8%
24%
13%
These figures clearly show that after processing the second preposition in the ambiguous
sentences, a brief pause is taken by nearly a quarter of the participants. This is three times
higher than the same prepositions in the unambiguous sentences and therefore shows the
difficulty faced by the subjects at this point in the sentences.
6.7 Main Clause Vs Complement Clause
This type of temporary ambiguity is typified by the employment of a past tense verb
directly following the subject of the sentence which is followed by a qualifying phrase that
constitutes the complement clause. The small alteration that was used to distinguish the
unambiguous sentence was the use of the words “that was” to denote the beginning of the
complement clause to the reader.
89
Figure 6.7.1
Main Clause Vs Complement Clause
Version 1 Version 2 Version 3 Version 4 Version 5 C
Subject 1
1
1
1
0
1
0
1
1
1
1
1
Subject 2
1
1
1
1
1
1
1
0
1
1
1
Subject 3
1
1
1
0
1
1
1
1
1
1
1
Subject 4
1
1
1
0
1
1
1
1
1
1
1
Subject 5
1
1
1
0
1
1
1
0
1
1
1
Subject 6
1
1
1
0
1
1
1
1
1
1
1
Subject 7
1
0
1
1
1
1
1
1
Subject 8
1
0
1
1
Subject 9
1
0
Subject 10
1
0
Figure 6.7.2
90
Main C Vs Comp C
Ambiguity control questions
Unambiguous
40
Ambiguous
38
26
30
Questions 20
12
10
0
0
Correct
Incorrect
The chart of figure 6.7.2 represents the data in figure 6.7.1. The first thing of note is
that the unambiguous sentences posed no difficulty as expected. About 33% of the subjects
however, subsequently failed to resolve the ambiguity satisfactorily in relation to the
ambiguous sentence. This suggests that this type of ambiguity is relatively difficult to
interpret correctly.
The temporal analyses of the sentences were studied (graphs 5V1A – 5V5B), once again,
in the area where pauses were expected in order to resolve the temporary ambiguity. For
this category, it is expected that the ambiguity will be present around the last few words in
the sentence, where the subject realizes that the clause headed by the past tense verb is in
fact a complement clause. This is illustrated below.
Unambiguous
16%
11%
3%
--------.
Ambiguous
|||
----13%
11%
16%
These percentages show that pauses were taken in the ambiguous sentences by
double the amount of subjects after the expected reanalysis denoted by the triple bar when
compared to those for the unambiguous sentences. Comparing the two shows that the
91
temporary ambiguity is not detected satisfactorily however due to the equal presence of
pauses all the way along. We can also see that the overall amount of pauses on the words in
the ambiguous sentences was higher than those without ambiguity. Despite this however,
the results do not show satisfactory evidence of the temporary ambiguity present, as the
same percentages for the initial word after the reanalysis were recorded.
6.8 Context Paragraphs
The results of the ambiguous sentences which appeared at the end of a paragraph are
treated somewhat differently to the other results. In this section, the data from the ambipairs are separated so that the data from the ambiguous part can be juxtaposed with the data
from the context ambiguous sentences. In figure 6.8.1, the percentage of control questions
that were answered correctly in each category, both for the isolated ambiguous sentences
and the context ambiguous sentences, can be seen.
Figure 6.8.1
Context Vs Isolated
Ambiguity control questions
Context
100
80
Percentage 60
Correct (%) 40
20
0
Comp
V Rel C
Obj NP
V MC
Sub
Isolated
Dir Obj
V
Embed
Sub
Prep
phrase
att
Main V
Comp C
Temporary Ambiguity Category
This chart shows that categories three and five proved to be the least difficult to resolve
when a context has been provided, 100% in each case. This is an improvement for category
five only. Category one and two are quite similar in success rate for the context sentences
but the resolution of category one ambiguity has decreased considerably. Category four was
the most difficult to resolve after a context, to the point where the overall resolution is
92
unsatisfactory. The respective time it took for the participants to answer the questions for
isolated and context sentences is described in the chart in figure 6.8.2 below.
Figure 6.8.2
Context Vs Isolated Respose Times
Context ambiguous
Isolated ambiguous
1200
1000
800
Time (Sec) 600
400
200
0
Comp Obj NP
V Rel C V M C
Sub
Dir Obj
V
Embed
Sub
Prep
Main V
phrase Comp C
att
Categories
The response times for category one and two are similar for both sets of sentences. The
time taken to answer the context question of category three was much higher than the
isolated version. The same can be said for category four. Overall however, category five
seems to have taken the least amount of time to answer for both sets of sentences. The
results for each of the context sentences in relation to the pauses taken in the selected region
of the sentences will now be discussed. The illustrations in the respective previous sections
are referred to.
Complement clause Vs Relative clause
The first category shows that the context ambiguous sentence showed a 10% pause
rate at the first and second words after the reading of the word “that”. This is significantly
lower than the figures for the isolated sentence.
Object noun phrase Vs Main clause subject
The second category’s context sentence figures merely show a 3% pause level after
processing the first word after the triple bar. The rest of the words did not show the
immediate presence of ambiguity resolution.
Direct object Vs Embedded subject
93
The figures for the context sentence of the third category show a 3% pause rate after
processing the first word of the selected area, another 3% over the word before the expected
re-analysis and 3% for the first word after the expected reanalysis. Once again, this does
not show any significant pauses in the sentences.
Prepositional phrase attachment
The context sentences of the fourth category only show a 3% pause rate over the
second preposition. The presence of any ambiguity resolution cannot be seen from these
results.
Main clause Vs Complement clause
The context sentence for the fifth category shows a 3% pause rate over the last two
words in the sentences as described above. Once again however, this is far less than any of
the isolated sentences.
6.9 Conclusion
In this chapter we have discussed the results to the experiment described in chapters four
and five. The results of the ambi-pairs have been discussed separately to the context
paragraphs. The ambi-pairs are compared within the confines of each ambiguity category
whereas the context sentences of the paragraphs are reported in comparison to the isolated
ambiguous sentences of the ambi-pairs. This is achieved in section 6.8 where the different
data sources are discussed. The results on the whole are quite satisfactory.
It remains to mention that seven of the participants guessed what the experiment was about
and out of these, six were linguists. This means that a very low percentage of the subjects
were aware of what was being tested and this upholds the prerequisite that the results reflect
the unconscious nature of the experiment. The fact that most of these were linguists implies
that past formal instruction in linguistics has a role in making the unconscious linguistic
processes, conscious ones. The next chapter will discuss the relevance of all the results and
the implications that can be made from them.
94
Chapter 7
Analysis of Results
95
7.1 Introduction
In this section the significance of the results will be discussed. There are several
topics on which the analysis of the results will focus. The first topic that will be addressed
is which ambiguity category caused the most difficulties to process and possible reasons as
to why this was the case. Section 7.3 deals with the analysis of the results with respect to
the two participating sexes. Section 7.4 treats the linguistic orientation aspect of the
experiment while section 7.5 looks into the results for the bilinguals that took part.
The penultimate section in the chapter delves back into the processing theories that
were discussed in the third chapter with a view to inferring possible conclusions from the
experiment at hand. Which processing theory is a more likely reality relies on the
significance of the data retrieved from the context sentences, as opposed to the isolated
ones. The chapter will end with a brief discussion of the experiment’s results.
7.2
Temporary Ambiguity Categories
An aim of the experiment was to study five different temporarily ambiguous
categories seeing which ones proved more difficult to process and offer answers as to why
this could be the case. Ambi-pairs were used to contrast the processing data between the
ambiguous and unambiguous sentences. Figure 7.2.1 shows the response times to both parts
of the ambi-pairs in each category.
Figure 7.2.1
96
Response times
Aggregate unambiguous time
700
600
500
Time (sec) 400
300
200
100
0
Aggregate ambiguous time
Comp Obj NP
V Rel C V M C
Sub
Dir Obj
V
Embed
Sub
Prep
Main V
phrase Comp C
att
Categories
What is apparent about the analysis of this graph is that the ambiguous sentence control
questions all required roughly the same amount of time for the subjects to answer,
regardless of the category. The Prepositional Phrase Attachment ambiguity records slightly
the longest time. Interestingly, category three, Direct Object Vs Embedded Subject
category, is the only category to receive more time for the unambiguous sentence control
questions to be answered than for the ambiguous ones. This may seem quite strange as the
opposite would be expected. The clarity of the sentences of this category must therefore be
better in the ambiguous part of the ambi-pairs. This implies that we may be more practiced
at interpreting the ambiguous construction, albeit, at a very slow pace.
Categories two and four, Object NP Vs Main Clause Subject and Prepositional
Phrase Attachment, both have the largest difference between the two parts of the ambipairs, the latter being the overall slower of the two. This would imply that these types of
ambiguous constructions are much less common and hence more difficulty to process. The
next graph, figure 7.2.2 investigates the correctness of the control questions.
Figure 7.2.2
97
Isolated Control Questions
Unambiguous
100
80
Percentage 60
correct (%) 40
20
0
Ambiguous
Comp Obj NP
V Rel C V M C
Sub
Dir Obj
V
Embed
Sub
Prep
Main V
phrase Comp C
att
Category
This shows that Direct Object Vs Embedded Subject sentences were resolved with an
excellent success rate even though the time it took for the subjects to respond was much
higher then the other categories. Both parts of the ambi-pairs were answered at nearly a
100% correct rating, once again it is the ambiguous control questions that were answered
slightly the better, a trend that is mimicked in the response times for that category.
Complement Clause Vs Relative Clause ambiguity shows that the ambiguous
sentences were resolved much better than the unambiguous ones even though it took about
the same amount of time to answer both. Once again, this is an unexpected result as you
would expect the ambiguous sentences to prove more difficult.
The greatest difference was recorded in Object NP Vs Main Clause Subject
category where the understanding of the ambiguous sentence proved extremely poor, less
than 50%, especially compared to its unambiguous 100%. This would confirm what its
response times show as its ambiguous sentence times are much greater.
Prepositional Phrase Attachment and Main Clause Vs Complement Clause have similar
differences between the two parts of the ambi-pairs. This difference is reflected in the
response times for both categories too with the prepositional phrase attachment seeming
slightly more difficult.
If we look at the temporal analysis of each category (appendix graphs), we can see that
Complement Clause Vs Relative Clause, Object NP Vs Main Clause Subject, and
Prepositional Phrase Attachment categories detect a significant pause percentage increase at
the expected re-analysis point of the ambiguous sentences. Interestingly, this is not shared
by the other two categories. With respect to category three, the temporary ambiguity
encountered in processing sentences of this kind proves least troublesome to the point
where it does not even show on the word to word analysis analysis. Category five shows a
little more anticipated pausing but still unsatisfactory.
98
In light of this analysis, category one seems to conform to the expected results of the
presence of temporary ambiguity. The temporal analysis showed the expected pauses were
present, the correctness of the ambiguous sentences was far less and the time it took to
answer the ambiguous sentences questions was much higher too. The second category
showed similar results for the anticipated pauses and also results for the questions that
conform to the presence of temporary ambiguity, at a moderately high difficulty level of
processing. Category three seems to be the easiest to process as least difficulty was
encountered. The ambiguous sentences actually proved to be both easier to understand and
quicker to answer questions on. There wasn’t even temporary ambiguity found through the
pauses in the processing of the sentences. This would suggest familiarity with constructions
of category three. Category four on the other hand seems to be the most difficult to process.
The pauses show a clear re-analysis area in the sentence, the ambiguous sentence questions
were much more poorly answered and the time it took to answer all leads to the conclusion
that this category proved the most difficult. This may be explained through the low
frequency of constructions in the English language of two prepositional phrases side by
side. Finally, category five demonstrated inconclusive pausing in the sentence to claim that
it is proof of temporary ambiguity. Despite this however, the level of correct answers and
response times would suggest that it is a relatively difficult ambiguity to process.
7.3 Male Vs Female
The sex of each participant is gathered in the final form that each subject had to fill
in. From the data gathered at this part of the experiment, a few charts and conclusions can
be drawn as to how each sex dealt with the temporary ambiguity. The sources of data that
are discussed in this section are the percentage of correct answers to the ambiguity control
questions and the response times to these questions. Figure 7.3.1 below shows the response
times for male and female participants.
Figure 7.3.1
Ambiguity Control Questions
Response Times
Unambiguous
Aggregate 100
Exposure time
Per person
(sec)
0
Unambiguous
Ambiguous
Ambiguous
Male
Female
59.6215
52.10805
77.02961111
62.36585
Sex
99
This chart shows that for both the ambiguous and unambiguous questions, the male
participants took longer to answer. In the case of the ambiguous sentences, it took them a
lot longer. The females answered the ambiguous questions nearly as fast as the males
answered the unambiguous questions. This suggests that women are able to resolve
temporary ambiguity faster, but in figure 7.3.2 below, the accuracy of these answers can be
seen.
Figure 7.3.2
Ambiguity Control Questions
Unambiguous
Ambiguous
100
Percentage
Correct (%)
50
0
Male
Female
Unambiguous
91.11111111
93
Ambiguous
71.11111111
74
Sex
These figures show that women’s answers were more accurate than the men’s. Taking
figure 7.3.1 into account, it can be concluded that overall female temporary ambiguity
resolution is more efficient than male resolution. The results from the experiment show that
women can resolve the individual constituents of temporarily ambiguous sentences both
faster and more accurately than men.
7.4 Linguists Vs Non-linguists
The next area of interest in relation to the results gained from the experiment is
whether or not linguists posses an enhanced ability to process and understand quirks in
language such as temporary ambiguity better. To investigate this further, the same charts
have been plotted as the last section. This will show us the speed and accuracy of the
linguist’s resolution.
Figure 7.4.1
100
Ambiguity Control Questions
Response Times
Unambiguous
Aggregate 100
Exposure time
Per person
(sec)
0
Ambiguous
Linguists
Non-linguists
Unambiguous 51.25864286 58.238625
Ambiguous
65.19092857 71.71570833
Linguistic Orientation
As one might expect, overall, the linguists that took part in the experiment answered both
the ambiguous and unambiguous sentences control questions faster than non-linguists.
Interestingly, the actual difference between the ambiguous and unambiguous parts were
quite similar which would suggest that despite the linguists being much faster, the actual
difficulties in resolving the ambiguity remained constant. The accuracy of the answers may
show more as to whether this might be the case.
Figure 7.4.2
Ambiguity Control Questions
Unambiguous
Ambiguous
100
Percentage
Correct (%)
50
0
Unambiguous
Ambiguous
Linguists
Non-linguists
100
87.5
80
68.33333333
Lintguistic Orientation
101
These figures confirm that linguists do possess a better understanding of language. They
failed to answer one unambiguous question wrong and scored extremely highly on the
ambiguous sentences. As you can see, the accuracy difference between the ambiguous and
unambiguous sentences for both categories remains constant. This suggests that having
studied linguists before improves accuracy and speed of temporary ambiguity resolution but
the fundamental challenges to each individual in processing the two types of sentences
remain the same regardless.
7.5 Bilinguals Vs Native speakers
The next analysis is that of the difficulties that a bilingual faces when confronted by
temporary ambiguity in a sentence. One would expect that a non-native speaker would find
linguistic pitfalls such as ambiguity an extremely difficult task. The speed and accuracy of
answers to the control questions were once again used to draw this contrast between the two
types of people who completed the experiment.
Figure 7.5.1
Ambiguity Control Questions
Response Times
Unambiguous
Aggregate 100
Exposure time
Per person
(sec)
0
Ambiguous
Bilinguals
Unambiguous 68.22228571
Ambiguous
Native
52.832
71.66285714 68.78096774
English Language
The chart demonstrates that, as predicted, native speakers can understand unambiguous
sentences much more quickly and the ambiguous ones a bit more quickly. Interestingly, the
bilingual results show that whether or not temporary ambiguity is present in the sentence
102
makes very little difference to the time it takes them to respond. The implication for this
would mean that foreign speakers of a language might not undergo the same sentence
processing procedure as they do in their own language. This would need to be investigated
further for a claim of such sort to be made but the evidence does point to this being the
case. The accuracy of the answers can be seen from figure 7.5.2.
Figure 7.5.2
Ambiguity Control Questions
Unambiguous
Ambiguous
100
Percentage
Correct (%)
50
0
Bilinguals
Native
Unambiguous
82.85714286
94.19354839
Ambiguous
57.14285714
76.12903226
English Language
The accuracy of the answers conforms to what one would expect for these two types of
people. Overall, native speakers find it very easy to resolve unambiguous sentences
whereas bilinguals find it more difficult. The similar response times for both sets of
sentences for bilinguals is not reflected however in the accuracy of their answers, the
ambiguous sentences proving a lot more difficult to extract the true meaning. On the whole,
these results would conform to those one would expect for a foreign speaker.
7.6
Processing theories revisited
The processing theories that were described at the end of chapter three will now be
discussed once again. This time it will be in light of the experiment and the results that have
been gained through the analysis of how humans process temporarily ambiguous sentences.
To briefly recap on what is the main topic of this section, we must summarise the two
theories once again.
The serial model suggested by Frazier&Fodor (1978) claims that the syntax faculty
in the brain resolves the constituents to a sentence before any semantic interpretation is
resolved. The parallel model on the other hand suggests that a semantic interpretation is
gained during the interaction of both the semantic and syntactic faculties of the brain. These
two models can be put to the test using the results from the current experiment.
103
If the serial model is a better representation of the human brain, then the existence
of prior contextually qualifying sentences would make a negligible difference. In this case,
the subjects would record similar results for both the isolated and context sentences of the
current experiment. If the parallel model is more representative of the processing procedure
of the brain then one would expect that the semantic mapping of previous sentences would
guide subjects in resolving the temporary ambiguity faster than if the ambiguous sentence
was isolated. This is due to the fact that if the subject already possesses a likely semantic
interpretation for the constituents of the sentence, he can apply the correct syntactic
structure to the temporarily ambiguous sentence faster as it is a more appropriate analysis.
All this relates to the current experiment where the context ambiguous sentences
can be contrasted with the isolated ambiguous sentences of the ambi-pairs. This is achieved
through the analysis of the response times and the accuracy of the answers to the ambiguity
control questions. Figure 7.7.1 shows the figures for the response times.
Figure 7.6.1
Context Vs Isolated Response Times
Series1
40
Average Time
20
Per Slide (sec)
0
context
Series1 34.3087556
isolated
27.8790922
Sentence orientation
These figures would suggest that prior semantic mappings have no effect on the resolution
of temporary ambiguity. It is clearly shown that, on average, it took 6.5 seconds longer for
each participant to answer the control questions when the sentence appeared in context. One
must concede however, that the results for the isolated sentences are far more representative
of the reality than the context results. This is due to the fact that there is five times the
amount of material to work on in relation to the isolated sentences as each subject sees five
examples of them compared to just one context sentence. Further material would be
necessary for a conclusion to be made. However, it does demonstrate an initial trend that
the resolution of temporary ambiguity is not aided by prior contextual knowledge that may
104
guide a person to the correct syntactic interpretation more quickly and hence, this would
favour the serial model.
Figure 7.6.2
Context Vs Isolated Respose Times
Context ambiguous
Isolated ambiguous
1200
1000
Aggregate 800
600
Time (Sec) 400
200
0
Comp Obj NP
V Rel C V M C
Sub
Dir Obj
V
Embed
Sub
Prep
Main V
phrase Comp C
att
Categories
Figure7.6.2 tells an interesting story. In only two of the ambiguity categories did the
context ambiguous sentences take longer to resolve. This would have the effect that the
figures for all the context sentences would raise. Clearly, this does not reflect the reality. It
has already been seen that isolated sentences in category four are the most difficult to
resolve and prior contextual information, it seems, just convolutes the overall analysis. If
categories three and five were omitted from the results, the opposite trend to that typified by
figure 7.6.1 would result. Interestingly, category five seems to be the easiest to resolve
when there is prior semantic knowledge which may guide the syntactic analysis. This
suggests a preference to the use of the parallel model.
Figure 7.6.3
Context Vs Isolated Response Times
Context
Isolated
30
Average Time
Per Slide
20
10
0
Male
Context 21.371556
Female
Linguists
Bilinguals
12.9372 16.765143 19.032571
Isolated 15.405922 12.47317 13.038186 14.332571
105
To help in the analysis of response times and possibly further refine a possible
conclusion, figure 7.6.3 compares the response times per slide of the different areas of
analysis. As the diagram shows, in every case, the context sentences proved longer to
resolve. In the case of the females, this length is negligible. The most surprising of these
results is the difficulty faced by the males in the experiment to resolve the context
sentences. The figures show that it even took bilinguals less time to resolve such linguistic
pitfalls. As the male figures seem a little unsatisfactory, a study was performed on the data
for the males. It was discovered that two males, one of them bilingual, took an aggregate of
126 seconds to answer their context sentence control question. This added more than 7
seconds to the male average per slide that can be seen in figure 7.6.3 and as such explains
the unrealistically high figures. The reason for such pauses might be explained by a mere
distraction in the room at the time of the experiment.
Taking the above observation into account, one can conclude that there is very little
difference between the times taken for isolated sentences and those in context. As stated,
more data is needed for the context sentences in order to make the results more conclusive.
We can however, take a look at the categories again to see if there is any trend between the
isolated ambiguous sentences and the context ambiguous sentences. This is achieved in
discussing whether the answers to the control questions were answered correctly or not.
The following diagram illustrates the trend.
Figure 7.6.4
Ambiguity Control Questions
Series1
80
60
Percentage
40
correct (%)
20
0
Series1
Context
Isolated
55.83333333
72.55555556
Once again, it can be seen that the context questions were much more poorly answered to
the point where nearly half of them were answered incorrectly. It is necessary to reiterate
that the lack of equal material may play a role in these figures but it does appear that
context plays little to no role in the resolution of temporary ambiguity. The next graph will
show if any ambiguity category in particular pushed this poor accuracy level up for the rest.
106
Figure 7.6.6
Context Vs Isolated
Ambiguity control questions
Context
100
80
Percentage 60
Correct (%) 40
20
0
Comp
V Rel C
Obj NP
V MC
Sub
Isolated
Dir Obj
V
Embed
Sub
Prep
phrase
att
Main V
Comp C
Temporary Ambiguity Category
As you can see, all the categories apart from the third and fifth decreased in the accuracy of
their context answers. Category four decreased to such an extent that one could say the
subjects resolved the ambiguity very poorly. The lack of data could also be a contributing
factor. Category one’s accuracy dropped by over a half when in context whereas category
two’s level of accuracy remained relatively constant. As for category three, it was equally
well answered regardless of its location. Category five is the only category of ambiguity
that recorded faster and more accurate answers to the control questions in context than the
isolated cases. This implies that prior semantic interpretations may guide a person to a
faster and more appropriate syntactic analysis in this type of temporary ambiguity. The
individual figures for the selected analyses may help to shed some light on the make up of
the above diagrams.
Figure 7.6.5
107
Ambiguity Control Questions
Context
Isolated
80
60
Percentage
40
Correct (%)
20
0
Male
Female
Context 66.666667
45
Isolated 71.111111
74
Linguists
Bilinguals
57.142857 57.142857
80
57.142857
The above chart shows that although the men were generally slower to answer the
questions, they were a lot more accurate in their answers to the context sentences than the
women. This suggests that women may have a poorer working memory. The context
sentences were unsatisfactorily answered overall by the females which would lead to
suggestions that they found resolving temporary ambiguity in context much harder.
Interestingly, the bilingual accuracy of answers did not change from isolation to context,
which suggests that they process each sentence as though it stands alone. Overall, it can be
concluded that every participant found it more difficult to resolve the ambiguity in context.
The results that have arisen from the experiment can be applied to the notion of which
processing theory seems the more likely. Although further research is needed, and more
material on which to base the results, the early indication is that context plays a negligible
role in the resolution of temporary ambiguity. In some categories it proved to make a
difference whereas in others it had the opposite effect. Taking everything into account, the
existence of context can be deemed, at least from the current experiment, that it has little
effect on the resolution of temporary ambiguity. This would lead to the conclusion that we
process sentences more akin to the serial model of chapter three. It must be stressed
however that the data for the context sentences is inadequate to firmly conclude this and as
such I will refrain from doing so. The study does however give an initial insight into the
negligible role context may play on the resolution of sentences. It suffices to mention that if
these results demonstrate the actual trend, we do, in fact, process sentences serially.
7.7
Discussion
This section will discuss all the relevant data that was gathered in relation to the two areas
of the experiment, the ambi-pairs and the context paragraphs. Final conclusions to the
experiment at hand will be offered.
108
7.7.1 The Ambi-pairs
The study of the ambi-pairs for each category generally demonstrated the presence
of temporary ambiguity through the analysis of the pauses. In every case apart from the
Direct Object Vs Embedded subject ambiguity this proved a success. A possible reason for
the lack of success in this category might be that we are well practiced at resolving such
constructions in everyday language. This is due to the fact that the verbs employed are
commonly found in a number of syntactic constructions and as a result, we are well
practiced at quickly reanalyzing them. The accuracy of answers to the control questions
was impeccable which would support such a claim. The fallacies noted in section 3.4 with
respect to embedded subjects did not reflect in the results implying that mature readers do
not suffer from early reading difficulties.
Interestingly, the Object Noun Phrase Vs Main Clause Subject ambiguity showed
that the percentage of pauses at the sight of a comma in the unambiguous sentences, was
equal to the percentage of pauses that were taken in order to resolve the relevant ambiguity.
This needs further investigation if there can be any claims made to the relevance of this
figure. The accuracy of the answers to the control questions for this category was
insufficient to claim that the ambiguity was satisfactorily resolved as there was a drop in
understanding by just less than 50% compared to the unambiguous sentences. This suggests
that the use of a comma to disambiguate the meaning of the sentence is extremely important
as we have very little practice at seeing such ambiguous sentences.
The results for the Prepositional Phrase Attachment ambiguity pauses showed three
times the amount of pauses on the second preposition, which would suggest a difficult
processing point in the sentences. Both sets of questions were answered satisfactorily with
the ambiguous ones recording a drop of less than 25% whilst maintaining an 80%
understanding rate. This suggests that although there are significant pauses during
processing, the syntactic reanalysis is good enough to resolve the ambiguity with success.
Results for the Main Clause Vs Complement Clause ambiguity showed mediocre
results in relation to the pauses picked up in the sentences. The resolution of ambiguity
however was sufficient to claim that the reanalysis functioned satisfactorily with a total of
just less than 80% of participants answering the ambiguous sentence control question
correct. This represented a drop of roughly 33%. Overall, this category was dealt with quite
easily by the subjects, which might reflect its frequency of use in everyday language.
The results for the Complement Clause Vs Relative Clause category showed an
interesting trend. It was the only category that had its ambiguous control questions
answered more accurately than the unambiguous ones, albeit a slight margin. A total of a
more than three-quarters resolved the ambiguity successfully. Pauses showed in the
expected areas as well which suggests that this type of ambiguity, whilst requiring a midsentence reanalysis, can be resolved easily enough.
109
7.7.2
The Context Paragraphs
The results for the context paragraphs showed a mixture of trends. In every case, the
pauses were not found to exist in any substantial way, which leaves the data on response
times and accuracy the only method of investigating whether the resolution of temporary
ambiguity is affected by prior context. In categories one, two and five, the response times to
the control questions for the context sentences were less than that of the isolated cases. This
might suggest the presence of a semantic interaction in their resolution.
The Object Noun Phrase Vs Main Clause Subject ambiguity showed negligible
differences between both the response times and accuracy of answers. The accuracy of the
context answers are slightly worse than in the isolated cases which could possibly be
explained by the marginally faster response times for these sentences.
The Direct Object Vs Embedded Subject ambiguity category showed similar
degrees of accuracy for both the isolated and context sentences. This was not reflected in
the response times for this category. The isolated sentences took a lot less time to respond
to, which suggests that although the ambiguity was successfully resolved in the context
sentences, prior semantic knowledge did not aid in the faster resolution of it.
Complement Clause Vs Relative Clause and Prepositional Phrase Attachment
ambiguity showed that when a context is provided, the resolution of the ambiguity is
insufficient. This processing difficulty is also reflected quite clearly in the much larger
response time for the latter category. Despite the first category having a slightly shorter
response time for the context sentences, the failure to answer the control questions with a
sufficient degree of accuracy means that these ambiguities do not show that prior contextual
knowledge aids in the ambiguity resolution.
Main Clause Vs Complement Clause ambiguity showed that the accuracy of
answers improved when a context was provided. This was the only category that
demonstrated that this increase in accuracy did not imply a larger response time. This
suggests that use of prior semantic knowledge in the processing of sentences may play a
heavy role in the resolution of Main Clause Vs Complement Clause ambiguity. More
research is necessary however if this claim is to be upheld.
The analysis of the genders showed that women resolve temporary ambiguity faster and
more accurately in the isolated cases but the opposite when there is a context present.
Linguists showed that formal instruction in the make-up of language helps when you
encounter such ambiguity. The bilinguals who took part in the experiment suggested that
they might process each sentence in isolation anyway so that prior context becomes
irrelevant. They naturally find ambiguous sentences more difficult in every sense.
In relation to the context analysis, category five seems to be the most affected by the
presence of context. The answers and speed of answers improved greatly where there was a
qualifying context. This is the only category however, that showed any signs that the
110
human processing faculty is informed by the interaction of semantic and syntactic
mappings. Due to the conflicting results for the context paragraphs, the debate must be left
open on whether our faculty is a serially or parallel based system.
Chapter 8
Conclusion
111
8.1 Introduction
This chapter is the conclusion to this paper. It will sum up all that has been gained from
the experiment to investigate temporary ambiguity resolution in humans. The particular
difficulties that were faced will be discussed and a suggestion of further study will bring the
chapter to a close.
8.2
Achievements and Knowledge Gained
The completion of this study has provided me with knowledge in several important areas
relating to computational and experimental linguistics. The first of these must be in relation
to ambiguity. Before undertaking this study, I was aware of this phenomenon but knew
very little about it. Through research in other works, see Crain&Stedman (1985), Forster
(1979), Kurztman (1984), Nive (1993) and Traxler (2002), I was able to grasp the uses of
ambiguity and what can be implied from the study of it.
To complement this area of study, sentence processing was another area that was heavily
researched, see Townsend&Beaver (2001), Merril Garett (1979), Forster (1979)
Frazier&Fodor (1978). This allowed me to understand all the principles that are involved in
how humans process sentences and more specifically, how they do this in the presence of
ambiguity and what processing ambiguity can reveal to us.
This leads on to the knowledge I gained on the experimental linguistics side. Once again,
research in the area, see Greenbaum (1977), Hoy (1999), Schütze (1996), allowed me to
construct a valid experiment to test the resolution of temporary ambiguity. To successfully
112
achieve this, the research I had done played a vital role. Information regarding parameters
and pitfalls to avoid were essential for me to achieve a valid experiment.
The computational aspect of the study involved the use of a number of bits of software.
Each one added to my knowledge of computer programs and how they can be used for
linguistic purposes. The web tool allowed me to set up an on-line experiment. ICECUP
allowed me to gather material for the experiment and Excel was essential for the results to
be analysed. Each program was used to its full potential and the knowledge gained from
these exercises will be invaluable.
In relation to the knowledge gained from the results, I have learned that temporary
ambiguity is a complex problem that can tell us a lot about the linguistic faculty of the
brain. There are different types of temporary ambiguity and the way in which we treat each
one is different. Some are easier dealt with than others. The study of the way we process
sentences is an area of deep complexity and it is of little wonder as to why there is still no
comprehensive agreement on a processing model.
8.3
Difficulties Encountered
There were several difficulties encountered throughout the study. The biggest difficulty
was in relation to the experimentation tool. A lot of time was spent constructing the
experiment as time and time again, the lack of its robustness caused problems. It also had to
be adapted slightly and although this was in the hands of the Experiment Manager, the
design of the experiment depended on it.
Learning to use three programs from scratch was also quite time consuming and added a
lot of difficulty to the overall study. The results that have been achieved from each one
however were astonishing and despite the difficulties in familiarising myself with them
initially, the rewards for doing so are evident.
8.4
Suggestions for Further Study
Suggestions for further study have been noted throughout the paper. The inconclusive
results for the relation that context plays on ambiguity resolution is the most obvious
catalyst for further studies. The investigation, through ambiguity resolution, of how we
process sentences is a topic that needs further attention. Each individual temporary
ambiguity category can also shed light on this and it is essential that each category be
studied in its entirety. Many conclusions can be drawn from results of ambiguity resolution
studies and it is this that must inspire the ultimate explanation of the human processing
faculty.
113
BIBLIOGRAPHY
Britannica Encyclopidia on-line
www.britannica.com
Caron, J., (1992). An introduction to psycholinguistics,
University of Poitiers
Chomsky, N., (1986). Knowledge of Language, Its Nature, Origin, and Use
New York: Praeger
Crain, S. & Steedman, M (1985). Natural Language Parsing: Psychological,
computational, and theoretical perspectives,
Cambridge University Press
Crain, S. & Steedman, M, (1985). On not being led up the garden path: The use of
context by the psychological parser
In D. Dowty, L. Kartunnen, & A. Zwicky (Eds.). Natural language processing.
NY: Cambridge University press
Forster, K.I., (1979). Levels of processing and the structure of the language processor
Monash University
In Holmes, V.M., (1979). Some hypotheses about Syntactic processing in sentence
comprehension
University of Melbourne
Frazier, L. & Fodor, J.D. (1978). The sausage machine: A new two-stage parsing model.
In Cognition, 6, 291-325
Geunnouni, M. (2000). Extension of a Web Based Environment for Cognitive
ScienceExperiments
Final year project
Greenbaum, S., Ed (1977a). Acceptability in Language
114
The Hague: Mouton
Greenbaum, S. (1977b). The linguist as experimenter
In Eckman, F., Ed., Current themes in linguistics: Bilingualism, Experimental
Linguistics, and Language Typologies
Washington: Hemisphere
Harrow, J (2004). Psycholinguistics Experiment into the grammaticality of complex
Sentences
Final year project
Howard S. Kurtman, Ambiguity resolution in the human syntactic parser: An
experimental study
Department of Psychology, MIT
Cambridge
Hoy, M.&Sheehan, K., (1999). Using e-mail to survey Internet users in the United
States: Methodology and assessment
In Journal of Computer-Mediated Communication
http:/www.ascusc.org/jcmc/vol4/issue3/sheehan3.html
Kenny, s., (1998), A Generic Automatic Experiment Creation and Presentation Tool
Final Year Project
Kurtzman, H.S., (1984). Studies in syntactic ambiguity resolution,
Ph.D. Dissertation, MIT
Nive, M., (1993). A computation model of syntactic processing: Ambiguity Resolution
from Interpretation
Dissertation: Unviersity of Pennsylvania
McGowan, C. (1999). Extension of a Web Based Environment for Cognitive Science
Experiments
Final year project
Merril Garett (1979). Sentence processing: Psycholinguistic studies presented by Merril
Garett
Lawrence Erlbaum Associates Publishers: Hillsdale, New Jersey
Ryan, N (2001). Web Based Experimentation in Cognitive Science
Final year project
Schutze (1996). The Empirical base of linguistics: Grammaticality, Judgements and
Linguistic Methodology
London: University of Chicago Press
Sentence comprehension: the connection to reading skills, May 2001
115
www.brainconnection.com
Sentence Structure, Michael Buckhoff
buckhoff.topcities.com
Searfoss, L.W., Readence, J.E., & Mallette, M.H. (2001). Helping Children Learn to Read,
Boston: Allyn & Bacon
Syntactic structures in music, Tom Sutcliff
www.harmony.org.uk
Townsend, D J., and Bever, T. G., (2001). Sentence Comprehension: The Integration of
Habits and Rules.
MIT Press
Traxler, M. J., (2002). Plausability and subcategorization preference in children’s
processing of temporarily ambiguous sentences: Evidence from self-paced reading
In The Quarterly Journal of Experimental Psychology, 2002, 55A(1), 75-96
http://ibs.derby.ac.uk/~Sigrid/language/TRAXLER.PDF
Whorf, B.L., (1950). Four Articles on Meta-linguistics
Washington: Foreign Services Institute
Wikipedia Encyclopidia on-line
en.wikipedia.org
116
Appendix
117
Experiment instructions
1.
2.
3.
4.
5.
You are about to read ten sentences and a paragraph
All the sentences are fully grammatical
The sentences and paragraph will be shown, word by word
To progress through the sentence, click the “next slide” button
After some sentences, you will need to answer a short question
Experiment sentences
Each subject (#1, #2, #3, #4, #5) sees one pair of ambiguous and unambiguous
sentences per category. These have been split up into five separate experiments as described
in chapter four. Each sentence has a #number to indicate which subject sees which
sentences and paragraphs.
118
Complement clause vs. Relative clause temporary ambiguity
1.
2.
#1,The spokesman told the journalist that the defendant annoyed that he should
go jump off a cliff.
#1,The spokesman told the journalist who was annoyed by the defendant that he
should go jump off a cliff.
Ambiguity question: who was told to jump off a cliff?
1. #2,The son argued with the workman that the children hindered that he was too old .
2. #2,The son argued with the workman who was hindered by the children that he was
too old .
Ambiguity question: who was too old?
1. #3,The old man suggested to the young lady that the men were courting that she
shouldn’t be so loving towards them.
2. #3,The old man suggested to the young lady who courted the men that she
shouldn’t be so loving towards them.
Ambiguity question: who shouldn’t be so loving?
1. #4,The boardroom discussed the presentation that the man gave that was shocking
2. #4,The boardroom discussed the presentation that was given by the man that was
shocking
Ambiguity question: who gave the presentation?
1. #5,The sergent persuaded the batallion that the king had lost that they had to fight
on.
2. #5,The sergent persuaded the batallion that had been lost by the king that they had
to fight on.
Ambiguity question: who had to fight on?
119
#2, The solidiers found themselves without a clue of where they were. They had been
seperated from their king and even their fellow comrades. The sergent finally came over
the brow of the hill. After a long deliberation, the sergent persuaded the battalion that
the king had lost that they had to fight on.
Object NP vs. main clause subject temporary ambiguity
1.
2.
#1,While Nigel was reading the newspaper lay unnoticed on the floor.
#1,While Nigel was reading, the newspaper lay unnoticed on the floor.
Ambiguity question: what was Nigel reading?
1. #2,Whilst Duncan was writing the manuscript fell to the ground.
2. #2,Whilst Duncan was writing, the manuscript fell to the ground.
Ambiguity question: what was Duncan writing?
1. #3,While Mary was mending the sock slipped off her leg.
2. #3,While Mary was mending, the sock slipped off her leg.
Ambiguity question: what was Mary mending?
1. #4,While Stevie was smoking the cigarette fell on his lap.
2. #4,While Stevie was smoking, the cigarette fell on his lap.
Ambiguity question: what was Stevie smoking?
1. #5,When the princess was sweeping up the dirt was everywhere.
2. #5,When the princess was sweeping up, the dirt was everywhere.
Ambiguity question: what was the princess sweeping up?
#3, The princess hated cleaning. She was always stuck with the worst jobs. The
children would always have the rooms back to their awful state by the time she was
finished. Before she started however, the rooms were in a worse state. For example,
when the princess was sweeping up the dirt was everywhere.
120
Direct Object vs. Embedded Subject ambiguity
1. #1,The smart student remembered the answer was not correct.
2. #1,The smart student remembered that the answer was not correct.
Ambiguity question: what did the student remember?
1. #2,The judge finally announced the punishment was insufficient.
2. #2,The judge finally annonced that the punishment was insufficient.
Ambiguity question: what did the judge announce?
1. #3,The prostitute warned the young boy was too small.
2. #3,The prostitute warned that the young boy was too small.
Ambiguity question: what did the prostitute warn against?
1.
2.
#4,The young woman suggested the way was over the hill.
#4,The young woman suggested that the way was over the hill.
Ambiguity question: what was suggested by the woman?
1. #5,The wise man finally believed the fortune teller was talking jiberish
2. #5,The wise man finally believed that the fortune teller was talking jiberish
Ambiguity question: what did the wise man believe?
#4, The wise man decided to see the fortune teller. He had been complaining to his wife
of a constantly reoccuring nightmare. His wife however was against the idea. She said
that the fortune teller is a load of nonsense. After an hour session with the fortune teller,
the wise man finally believed the fortune teller was talking jiberish.
121
Prepositional phrase attachment ambiguity: NP-modifier vs. VP-modifier
or VP-argument
1. #1,The cook placed the cake in the oven on the table
2. #1,The cook placed the cake that was in the oven on the table
Ambiguity question: where was the cake before it was moved?
1. #2,The princess put the frog on the basket on the table
2. #2,The princess put the frog that was on the basket on the table
Ambiguity question: where was the frog before it was moved?
1. #3,The soccer player kicked the ball on the penalty spot into the goal
2.
#3,The soccer player kicked the ball that was on the penalty spot into the goal
Ambiguity question: where was the ball before it was kicked?
1. #4,The mechanic finally attached the bolt on the bonnet to the car.
2. #4,The mechanic finally attached the bolt that was on the bonnet to the car.
Ambiguity question: Where was the bolt before it was attached?
1. #5,The musician plucked the harp in his hands on stage
2. #5,The musician plucked the harp that was in his hands on stage
Ambiguity question: where was the harp?
#5, It had been a hard days work for the mechanic. He only had a few more things to do
before he was able to go home. There were nuts and bolts everywhere. His last job was
always the hardest. The mechanic finally attached the bolt on the bonnet to the car.
Main clause Vs complement clause temporary ambiguity
122
1.
2.
#1,The boat floated down the river sank without a trace
#1,The boat that was floated down the river sank without a trace
Ambiguity question: what initially happened to the boat?
1.
2.
#2,The man gave the dog a bone fell on a bath
#2,The man gave the dog that a bone fell on a bath
Ambiguity question: what did the man give the dog?
1.
2.
#3,The horse raced past the barn fell
#3,The horse that was raced past the barn fell
Ambiguity question: what was happening to the horse before it fell?
1.
2.
#5,The speeding car driven past the police crashed.
#5,The speeding car that was driven past the police crashed.
Ambiguity question: What was the car doing before it crashed?
1.
2.
#4,The students taught in the open air do well
#4,The students who are taught in the open air do well
Ambiguity question: where do the students do well?
#1, A range of experiments were tried by the postgraduates. Students were taught in a
variety of different places to see how they would react to the environment. One conclusion
at the end of the experiment was, the students taught in the open air do well.
123
Temporary Ambiguity Results
1.
Complement clause Vs Relative clause
2.
Object noun phrase Vs Main clause subject
3.
Direct object Vs Embedded subject
4.
Prepositional phrase attachment
5.
Main clause Vs Complement clause
Bilingual and Linguist participation
Version 1
Version 1
- Subject 3
- Subject 4
Linguist
Linguist, Greek
Version 2
Version 2
Version 2
- Subject 5
- Subject 8
- Subject 10
Spanish
Linguist, Finnish
Linguist
Version 3
Version 3
Version 3
Version 3
- Subject 3
- Subject 4
- Subject 5
- Subject 6
Linguist, Greek
Linguist
Linguist
Linguist
Version 4
Version 4
Version 4
- Subject 3
- Subject 4
- Subject 5
Linguist
Linguist, German
Linguist, French
Version 5
Version 5
Version 5
- Subject 3
- Subject 5
- Subject 6
Linguist
Linguist, German
Linguist
1V1A
124
Comp C Vs Rel C (Version 1)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 1
a
off
cliff
go
Participants
Subject 1
jump
he
should
defendant
Words
that
by
Subject 3
the
annoyed
who
journalist
was
told
Subject 5
the
The
spokesman
Instructions
0
Instructions
The
spokesman
told
the
journalist
w ho
w as
annoyed
by
the
defendant
that
he
should
go
jump
off
a
cliff
18
5.75
3.25
1.032
1.391
0.671
0.578
0.578
0.718
0.688
1.141
2.781
0.891
0.703
0.625
0.75
0.891
0.625
0.579
13.656
11.12
1.461
1.446
0.651
0.609
1.224
0.706
0.728
1.072
1.394
0.814
3.679
1.265
6.602
0.65
0.688
0.721
0.625
0.521
14.22
Subject 3
6.626
1.215
0.349
0.403
0.465
0.362
0.449
0.496
0.45
0.491
0.401
0.719
0.48
0.459
0.593
0.497
0.482
0.474
0.483
20.903
Subject 4
30.756
2.677
1.301
1.06
0.976
0.821
0.784
0.968
0.773
0.89
0.937
0.801
0.8
0.641
0.746
0.594
0.688
0.578
0.719
24.904
Subject 5
21.036
11.697
2.392
1.01
0.655
0.843
0.681
0.682
0.659
0.593
0.864
0.687
0.715
0.695
0.843
0.645
0.71
0.528
0.495
18.796
Subject 6
84.825
4.29
2.627
1.708
1.585
1.976
1.41
1.085
2.794
1.035
1.099
6.127
2.196
0.447
3.935
11.759
0.748
0.61
0.59
28.641
Subject 2
1V1B
Comp C Vs Rel C (Version 1, Amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
Subject 5
Subject 4
Subject 3
Subject 2
a
off
Subject 1
cliff
Words
jump
go
should
Participants
he
that
annoyed
defendant
the
that
journalist
told
the
The
spokesman
0
The
spokesman
told
the
journalist
that
the
defendant
annoyed
that
he
should
go
jump
off
a
cliff
Subject 1
1.219
0.406
0.5
0.468
0.453
0.484
0.437
0.359
0.422
0.531
2.125
1.282
0.594
0.453
0.328
0.297
11.891
Subject 2
0.961
0.536
0.506
0.452
0.571
0.515
1.124
0.6
0.57
4.321
0.515
0.5
0.531
1.61
0.804
0.705
26.896
Subject 3
1.051
0.34
0.521
0.428
0.42
0.444
0.46
0.476
0.427
0.504
0.883
0.66
0.953
2.346
0.507
0.391
9.991
Subject 4
1.913
0.679
0.675
0.603
0.843
1.22
0.721
1.378
1.907
0.777
0.98
1.259
0.748
0.718
0.651
0.657
11.228
Subject 5
1.197
0.411
0.408
0.378
0.415
0.462
0.441
1.106
0.921
1.075
0.435
0.393
0.403
0.41
0.409
0.418
13.127
Subject 6
2.922
0.771
4.098
0.584
0.617
0.481
0.5
1.434
0.658
1.819
3.315
0.493
0.525
1.332
0.53
0.542
24.045
1V2A
125
Comp C Vs Rel C (Version 2)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
5
Subject 9
0
Words
too
old
he
Subject 1
was
that
children
by
the
hindered
who
Participants
Subject 3
was
workman
the
with
argued
The
son
Instructions
Subject 7
Subject 5
Instructions
The
son
argued
w ith
the
w orkman
w ho
w as
hindered
by
the
children
that
he
w as
too
old
Subject 1
22.722
22.062
1.903
1.432
1.452
1.172
0.911
0.771
0.821
1.412
0.751
0.601
0.971
0.701
0.951
0.661
0.821
29.692
Subject 2
1.674
1.32
0.445
0.415
0.415
0.307
0.538
0.568
0.4
0.43
0.338
0.399
0.492
0.522
0.446
0.445
0.492
8.952
Subject 3
17.459
1.702
0.952
0.594
0.594
0.547
0.719
0.547
0.531
0.625
0.594
0.546
0.594
0.625
0.61
0.733
0.625
10.978
Subject 4
22.966
2.063
2.082
1.382
0.971
1.031
0.821
1.762
3.674
0.981
0.711
1.372
0.901
0.981
0.881
2.303
0.76
9.261
Subject 5
31.96
2.436
0.812
0.594
0.578
0.562
0.516
0.531
0.484
0.469
0.5
0.546
0.484
0.5
0.469
0.422
0.437
11.794
Subject 6
24.172
4.656
1.406
0.766
1.406
0.656
1.625
0.719
1.266
0.828
0.641
0.547
2.422
1.172
0.547
0.813
0.625
16.484
Subject 7
19.047
2.765
0.985
0.828
0.828
0.578
0.625
0.641
1.203
0.657
0.625
0.578
0.5
0.453
0.547
0.406
0.484
8.733
Subject 8
18.216
4.337
1.972
1.372
1.732
1.195
1.829
1.612
1.188
1.42
1.395
1.044
1.661
4.69
2.356
1.11
0.876
Subject 9
13.948
1.832
0.591
0.711
0.571
0.54
0.441
0.671
0.5
0.541
0.511
0.39
0.2
0.651
0.281
0.39
0.861
8.952
Subject 10
6.88
1.673
0.961
0.651
0.761
0.491
0.561
0.501
0.45
0.511
0.541
0.46
0.701
0.451
0.661
0.591
0.901
11.767
8.904
1V2B
Comp C Vs Rel C (Version 2, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
5
Subject 9
Subject 7
The
son
Participants
too
old
Subject 1
was
he
that
hindered
the
Words
children
Subject 3
that
workman
the
with
argued
Subject 5
son
The
0
argued
w ith
the
w orkman
that
the
children
hindered
that
he
w as
Subject 1
2.624
0.56
0.61
0.491
0.581
0.561
2.053
0.751
0.731
0.611
0.511
0.54
2.214
0.61
6.991
Subject 2
1.225
0.543
0.729
0.604
0.589
0.527
0.48
0.481
0.682
0.543
0.512
0.527
0.496
0.528
too
13.894
old
Subject 3
1.296
0.484
0.422
0.453
0.453
0.468
0.594
0.516
0.515
0.593
0.531
1.171
0.812
0.453
9.869
Subject 4
4.615
1.092
0.631
0.53
1.192
0.721
0.721
0.691
0.64
0.531
0.711
0.621
1.382
0.59
27.525
Subject 5
3.92
0.407
0.421
0.453
0.468
0.422
0.484
0.407
0.406
0.515
0.562
0.687
0.453
0.422
8.795
Subject 6
0.968
0.469
0.454
0.593
0.625
0.875
0.766
5.36
0.485
3.359
0.453
0.64
0.484
0.453
21.156
Subject 7
1.312
0.484
0.468
0.5
0.452
0.453
0.515
0.531
0.5
0.499
0.468
0.671
0.515
0.546
9.04
Subject 8
1.307
0.837
2.317
0.669
0.685
1.188
0.693
2.3
6.294
4.004
1.632
0.952
1.241
0.901
20.092
Subject 9
1.071
0.411
0.31
0.281
0.44
0.361
0.23
0.26
0.211
0.27
0.39
0.371
0.691
0.42
5.657
Subject 10
1.342
1.803
0.731
0.531
0.46
0.511
0.701
0.561
0.551
0.681
0.581
0.661
0.6
0.671
12.688
1V3A
126
Comp C Vs Rel C (Version 3)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
0
Participants
them
loving
be
so
Subject 1
towards
Words
shouldn
that
she
the
men
courted
lady
who
young
the
to
suggested
man
old
The
Subject 7
Subject 5
Subject 3
The
old
man
suggested
to
the
young
lady
w ho
courted
the
men
that
she
shouldn
so
loving
tow ards
them
Subject 1
0.922
0.875
0.515
0.766
0.547
0.5
0.468
0.5
0.718
0.703
0.781
0.562
0.547
0.562
0.562
0.532
0.5
0.422
0.563
6.906
Subject 2
0.872
0.53
0.411
0.491
0.52
0.541
0.351
0.931
0.511
0.4
0.771
4.657
0.531
0.451
0.48
0.491
be
0.501
0.651
0.52
9.925
Subject 3
0.905
0.178
0.326
0.266
0.231
0.251
0.179
0.211
0.426
0.307
0.404
0.305
0.291
0.267
0.26
0.372
0.203
0.318
0.427
12.117
Subject 4
1.01
0.26
0.236
0.188
0.165
0.282
0.147
0.299
0.892
0.188
0.164
0.283
0.221
0.269
0.212
0.202
0.197
0.228
0.403
7.568
Subject 5
1.258
0.537
0.538
0.477
0.099
1.885
0.46
0.116
1.827
3.363
0.557
1.029
0.618
0.628
8.486
0.485
0.459
0.701
0.58
6.567
Subject 6
2.983
0.46
0.459
0.476
0.442
0.364
0.347
0.34
0.372
0.373
0.413
1.357
0.404
0.402
0.387
0.411
0.41
0.412
0.364
4.742
Subject 7
1.218
0.422
0.421
0.469
0.454
0.438
0.484
0.484
0.609
0.625
0.625
0.516
0.5
0.469
0.484
0.5
0.468
0.438
0.453
5.765
1V3B
Comp C Vs Rel C (Version 3, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
them
towards
so
be
she
Participants
Subject 1
loving
Words
shouldn
that
Subject 3
courting
were
the
men
that
lady
the
Subject 5
young
to
man
suggested
Subject 7
old
The
0
The
old
man
suggested
to
the
young
lady
that
the
men
w ere
courting
that
she
be
so
loving
Subject 1
1.437
0.546
0.532
1.016
0.922
0.953
0.766
0.703
0.703
0.828
0.906
0.766
0.828
1.906
0.594
0.578
0.531
0.578
0.937
0.75
9.452
Subject 2
1.693
3.966
3.134
4.557
1.922
2.925
3.044
2.624
2.514
2.563
2.273
3.686
6.509
3.305
2.393
shouldn
4.757
2.334
2.253
2.954
tow ards
3.996
49.261
them
Subject 3
1.018
0.52
0.431
0.236
0.418
0.332
0.313
0.289
0.371
0.753
2.308
0.328
0.297
0.418
0.39
1.882
0.435
0.385
0.435
0.452
14.94
Subject 4
1.522
0.356
0.242
0.474
0.891
0.22
0.284
0.238
0.252
0.778
0.19
0.227
0.485
0.219
3.342
3.68
0.547
0.507
0.418
0.396
13.868
Subject 5
1.203
0.483
0.461
0.562
0.539
0.515
0.475
0.511
0.504
0.506
0.569
0.531
0.882
0.931
0.85
2.73
0.708
0.619
0.706
0.755
16.277
Subject 6
1.003
0.447
0.395
0.399
0.403
0.443
0.411
0.42
0.386
0.444
0.405
0.419
0.403
0.442
1.118
0.435
0.42
0.435
0.434
0.46
6.835
Subject 7
1.313
0.906
0.921
0.719
0.766
0.75
0.672
0.735
0.688
0.781
0.687
0.656
0.671
1.375
0.875
0.672
0.719
0.797
0.75
0.64
6.672
1V4A
127
Comp C Vs Rel C (Version 4)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
Subject 4
Subject 3
presentation
was
shocking
man
that
by
the
was
Subject 1
Words
the
Participants
Subject 2
given
that
presentation
the
discussed
The
boardroom
0
The
boardroom
discussed
that
w as
given
by
the
man
Subject 1
2.406
0.797
0.734
0.64
1.5
2
0.734
0.546
0.547
0.454
3.375
0.75
0.75
23.75
Subject 2
1.128
0.533
0.533
0.517
0.69
0.517
0.658
0.643
0.579
0.548
0.533
0.486
that
0.485
w as
shocking
17.215
Subject 3
1.402
0.803
0.619
0.82
0.716
0.509
0.515
1.218
0.563
0.667
0.783
0.595
0.447
13.767
Subject 4
0.981
0.381
0.49
0.341
0.46
0.451
0.351
0.42
0.431
0.34
0.371
0.38
0.371
5.738
Subject 5
2.864
1.482
2.364
1.211
1.151
0.932
0.62
0.862
0.931
1.182
0.971
1.002
0.871
29.342
Subject 6
2.985
0.829
0.718
0.782
0.875
0.75
0.703
0.828
0.937
1.032
1.843
0.938
0.687
14.454
Subject 7
1.125
0.672
1.64
1.234
0.922
0.609
0.704
1.609
0.687
0.438
0.86
0.609
0.594
21.234
1V4B
Comp C Vs Rel C (Version 4, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
0
was
shocking
that
man
gave
Subject 1
Words
The
Participants
Subject 2
the
that
Subject 3
presentation
the
boardroom
Subject 4
discussed
The
Subject 5
boardroom
discussed
the
presentation
the
man
gave
that
w as
shocking
Subject 1
1.594
0.75
0.641
0.625
1.406
0.89
0.64
0.953
1.078
0.985
0.75
7.203
Subject 2
0.984
0.523
0.468
1.148
0.585
0.586
that
0.498
0.426
0.478
0.997
0.473
5.095
Subject 3
0.984
0.523
0.468
1.148
0.585
0.586
0.498
0.426
0.478
0.997
0.473
Subject 4
0.891
0.2
0.381
0.15
0.19
0.241
0.2
0.2
0.211
0.21
0.891
8.883
Subject 5
1.913
0.641
0.721
0.781
1.582
0.721
0.541
0.841
0.972
2.463
0.591
17.475
Subject 6
2
0.609
0.5
0.594
0.922
0.641
0.546
0.485
0.531
0.578
0.641
11.375
Subject 7
1.11
0.468
0.657
1.062
1.36
1.453
1.563
0.453
0.969
0.844
0.547
7.375
5.095
1V5A
128
Comp C Vs Rel C (Version 5)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
0
to
Words
on
fight
they
Participants
Subject 1
had
king
that
by
the
been
lost
that
had
the
batallion
persuaded
The
sergent
Subject 7
Subject 5
Subject 3
The
sergent
persuaded
the
batallion
that
had
been
lost
by
the
king
that
they
had
to
f ight
on
Subject 1
0.875
0.422
0.422
0.422
0.468
0.516
0.5
0.859
0.594
0.516
0.453
0.406
0.406
0.297
0.453
0.516
0.375
7.251
Subject 2
1.156
0.453
0.454
0.5
0.625
0.453
0.375
1.969
0.781
0.5
0.469
0.563
0.531
0.625
0.953
0.516
0.531
5.484
Subject 3
0.925
0.47
0.423
0.408
0.454
0.533
3.152
0.564
2.023
0.705
0.424
0.47
0.471
0.408
0.439
0.423
0.361
4.281
Subject 4
0.759
0.668
0.659
0.459
0.539
0.669
1.028
0.509
1.736
1.418
0.798
1.767
0.588
0.489
0.549
0.509
0.659
Subject 5
0.996
0.492
0.309
0.332
0.389
0.389
0.396
0.54
0.394
0.442
0.419
0.469
0.528
0.517
0.493
1.22
0.532
Subject 6
1.011
0.402
0.387
0.747
0.659
1.364
0.661
0.563
0.519
0.698
0.507
0.97
0.522
0.564
0.563
0.477
0.435
9.987
Subject 7
1.781
0.906
1.61
0.594
0.5
0.562
0.562
0.657
0.718
1.266
0.844
4.453
0.641
0.625
0.61
0.671
0.688
30.088
Subject 8
1.905
0.562
0.547
0.5
0.765
0.859
0.515
0.562
0.781
0.578
0.656
0.562
0.718
0.593
0.609
0.594
0.578
6.277
7.565
5.112
1V5B
Comp C Vs Rel C (Version 5, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
Subject 8
Subject 7
Subject 6
Subject 5
Subject 4
Subject 3
Subject 2
Subject 1
on
Words
fight
to
had
they
lost
that
Participants
had
king
that
the
batallion
the
sergent
persuaded
The
0
The
sergent
persuaded
the
batallion
that
the
king
had
lost
that
they
had
to
f ight
on
Subject 1
1.5
0.375
0.672
0.438
0.437
0.485
0.406
0.329
0.39
0.406
0.266
0.375
1.891
0.515
0.625
12.266
Subject 2
1.562
2.109
1.312
1.359
1.14
1.312
1.312
0.984
0.703
1.094
2.063
1.703
1.688
0.938
1.344
7.984
Subject 3
1.23
0.425
0.442
0.441
0.379
0.473
0.441
0.458
0.489
0.473
0.489
4.873
0.536
0.489
0.442
30.298
Subject 4
1.008
0.948
0.778
0.579
0.689
0.639
0.608
0.639
0.609
0.728
4.442
1.866
0.529
0.489
0.439
9.512
Subject 5
1.7
0.514
0.931
0.453
0.485
0.396
0.437
0.453
0.381
0.38
0.476
0.459
0.82
0.588
0.459
7.841
Subject 6
2.641
0.779
0.857
0.592
0.684
0.704
0.718
0.768
0.729
0.873
0.543
1.466
1.001
0.616
0.707
Subject 7
7.671
0.734
0.953
0.735
1.141
0.703
1.109
0.735
0.671
3.015
3.406
1.25
0.765
0.765
2.469
10.89
Subject 8
2.295
1.624
0.921
0.719
1.156
0.859
0.594
0.64
0.547
1.264
0.594
1.186
0.609
0.625
0.78
18.615
12.01
1C
129
Context Comp C Vs Rel C (Version 2, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
5
Subject 9
Subject 7
Words
Participants
on
to
Subject 1
fight
they
had
lost
that
king
Subject 3
had
that
the
the
battalion
persuaded
Subject 5
sergent
the
0
the
sergent
persuaded
the
battalion
that
the
king
had
lost
that
they
had
to
fight
Subject 1
0.531
0.511
0.561
0.55
0.501
0.561
0.601
0.611
0.46
0.501
0.491
0.631
1.101
0.571
0.721
6.67
Subject 2
0.248
0.311
0.295
0.373
0.388
0.342
0.295
0.28
0.279
0.28
0.264
0.264
0.497
0.435
0.295
8.119
on
Subject 3
0.468
0.406
0.562
0.438
0.375
0.453
0.499
0.343
0.422
0.437
0.36
0.359
0.406
0.484
0.469
8.917
Subject 4
0.491
0.47
0.571
0.541
0.49
0.521
0.551
0.55
0.421
0.46
0.451
0.731
0.741
0.611
0.721
16.641
Subject 5
0.438
0.422
0.781
2.202
0.453
0.453
0.484
0.437
0.516
0.343
0.453
0.453
0.594
0.546
0.406
Subject 6
0.547
0.515
0.641
0.5
0.859
0.625
0.547
0.672
0.515
0.64
0.719
2.391
0.781
0.968
2.062
16.75
Subject 7
0.484
0.5
0.562
0.422
0.422
0.453
0.39
0.422
0.437
0.406
0.422
0.641
0.468
0.452
0.453
13.615
Subject 8
0.723
0.891
0.86
0.701
1.196
1.053
0.693
0.773
0.727
0.914
1.101
1.228
0.836
0.749
1.029
8.031
Subject 9
0.221
0.23
0.21
0.24
0.26
0.32
0.231
0.27
0.2
0.23
0.261
0.27
0.351
0.45
0.26
6.208
Subject 10
0.471
0.431
0.52
0.511
0.461
0.46
0.501
0.481
0.461
0.45
0.521
0.571
0.531
0.53
0.471
5.949
8.779
2V1A
130
Obj Np Vs Main C Sub (Version 1)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
Subject 5
0
Subject 4
While
Nigel
Participants
Subject 3
w as
reading*
the
Subject 2
new spaper
Words
lay
unnoticed
Subject 1
on
the
floor
While
Nigel
w as
reading*
the
new spaper
lay
unnoticed
on
the
floor
Subject 1
0.922
0.343
0.297
0.265
0.281
0.266
0.516
1.078
0.5
0.656
14.953
Subject 2
15.716
1.283
0.523
5.323
0.53
0.513
0.466
0.643
0.503
0.417
6.396
Subject 3
1.951
0.306
0.428
0.396
0.459
0.371
0.46
1.122
0.46
0.451
6.399
Subject 4
1.178
0.731
0.586
0.682
0.72
0.571
0.626
0.746
0.625
0.637
9.421
Subject 5
1.075
0.466
0.378
0.393
0.44
0.706
0.626
0.443
0.458
0.466
6.277
Subject 6
2.367
0.7
0.588
0.878
0.603
0.601
0.651
0.836
0.635
0.452
9.699
2V1B
Obj NP Vs Main C Sub (Version 1, Amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
Subject 5
0
Subject 4
While
Nigel
Participants
Subject 3
w as
reading
the
Subject 2
new spaper
Words
lay
unnoticed
Subject 1
on
the
floor
While
Nigel
w as
reading
the
new spaper
lay
unnoticed
on
the
Subject 1
3.859
0.953
0.797
0.594
0.515
0.844
0.782
2.406
0.891
0.672
8.14
Subject 2
1.412
0.354
0.414
0.427
0.408
0.379
0.45
0.946
0.715
0.482
28.041
floor
Subject 3
1.766
1.408
1.181
0.943
0.725
0.759
1.979
1.511
0.784
0.883
12.269
Subject 4
1.631
0.671
0.534
0.468
0.478
0.586
0.569
0.848
0.604
0.71
12.099
Subject 5
2.935
0.93
0.609
0.543
0.596
1.629
1.291
1.563
1.035
0.604
16.907
Subject 6
2.935
0.93
0.609
0.543
0.596
1.629
1.291
1.563
1.035
0.604
16.907
2V2A
131
Obj NP Vs Main C (Version 2)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
Subject 10
Subject 9
Subject 8
Subject 7
Subject 6
Subject 5
Subject 4
Subject 3
5
Participants
fell
the
ground
Words
Subject 1
to
the
manuscript
was
Subject 2
writing*
Whilst
Duncan
0
Whilst
Duncan
w as
w riting*
the
manuscript
f ell
to
the
ground
Subject 1
1.682
0.671
0.501
0.721
0.601
0.551
0.531
0.56
0.561
12.258
Subject 2
1.241
0.512
0.449
0.481
0.59
1.039
0.573
0.543
0.465
10.529
Subject 3
1.109
0.531
0.687
0.624
0.625
0.546
0.531
0.485
0.562
8.62
Subject 4
1.752
1.041
0.861
1.702
0.801
0.691
0.541
0.471
0.42
5.117
Subject 5
1.453
5.873
0.734
0.797
1.468
0.5
0.468
0.562
0.469
7.295
Subject 6
1.25
0.812
0.547
0.5
0.625
0.875
0.531
0.516
0.468
Subject 7
2.124
0.546
0.437
0.453
0.453
0.499
0.5
0.874
0.562
10.07
Subject 8
1.486
1.348
1.053
1.244
1.017
1.333
1.029
0.861
0.773
12.867
Subject 9
1.662
0.451
0.32
0.331
0.3
0.271
0.26
0.2
0.26
6.228
Subject 10
1.272
0.58
0.561
0.801
0.591
0.691
0.751
0.601
0.541
6.409
8.703
2V2B
Obj NP Vs Main C sub (Version 2, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
Subject 10
Subject 9
Subject 8
Subject 7
Subject 6
Subject 5
Subject 4
Subject 3
5
Participants
ground
to
Subject 1
the
fell
manuscript
Words
Subject 2
the
writing
was
Duncan
Whilst
0
Whilst
Duncan
w as
w riting
the
manuscript
fell
to
the
ground
Subject 1
1.282
0.611
0.51
0.501
0.481
0.43
0.471
0.441
0.45
17.195
Subject 2
1.336
0.419
0.388
0.342
0.31
0.295
0.481
0.341
0.326
Subject 3
1
0.328
0.375
0.359
0.406
0.375
0.391
0.625
0.531
8.823
Subject 4
10.183
0.671
0.491
0.45
0.471
0.531
0.58
0.68
0.401
18.513
Subject 5
1.234
0.406
0.437
0.422
0.437
0.313
0.5
0.484
0.547
7.295
Subject 6
1.578
0.704
0.594
0.547
0.5
0.703
0.641
0.843
0.578
19.843
Subject 7
1.28
0.546
0.843
0.625
0.578
0.749
2.201
0.453
0.437
17.721
Subject 8
1.613
0.714
0.797
1.082
0.813
0.877
0.733
1.051
0.837
21.582
Subject 9
1.142
0.24
0.23
0.201
0.27
0.52
0.36
1.011
0.451
12.225
Subject 10
1.312
0.441
0.41
0.341
0.411
0.34
0.391
0.56
1.042
8.322
8.461
2V3A
132
Obj NP Vs Main C sub (Version 3)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
Subject 4
While
Mary
Subject 3
w as
the
Words
w as
Participants
Subject 2
mending*
sock
slipped
Subject 1
of f
her
leg
While
Mary
mending*
the
sock
slipped
off
her
leg
Subject 1
1.234
0.672
1.14
2
1.969
1.062
1.218
0.812
0.641
12.093
Subject 2
1.362
0.671
0.671
8.782
0.771
0.491
6.9
0.491
0.471
13.008
Subject 3
1.067
0.25
0.311
0.195
0.183
0.523
0.303
1.921
1.651
10.054
Subject 4
1.588
0.45
0.187
0.298
0.22
0.419
1.286
0.172
0.155
8.008
Subject 5
1.305
0.777
0.762
1.78
0.633
0.691
1.389
0.59
0.672
4.593
Subject 6
0.884
0.427
0.404
0.45
0.424
0.49
0.442
0.463
0.571
4.95
Subject 7
2.484
0.578
0.75
0.594
0.657
0.609
0.765
0.594
0.61
4.375
2V3B
Obj NP Vs Main C sub (Version 3, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
Subject 4
While
Mary
Subject 3
w as
mending
Participants
Subject 2
the
Words
sock
slipped
Subject 1
off
her
leg
While
Mary
w as
mending
the
sock
slipped
off
her
leg
Subject 1
1.031
0.484
0.484
0.64
0.532
0.5
1.031
0.891
0.547
12.812
Subject 2
0.891
0.411
0.4
0.411
0.601
1.101
1.883
0.601
0.461
7.45
Subject 3
0.738
0.253
0.221
0.268
0.244
0.172
0.341
2.258
0.26
15.631
Subject 4
1.018
0.195
0.18
0.172
0.148
0.26
0.146
0.148
0.404
13.704
Subject 5
1.538
1.065
0.499
0.548
0.509
0.477
0.469
0.437
0.434
7.631
Subject 6
0.89
0.405
0.404
0.387
0.422
0.397
0.453
0.404
0.469
10.664
Subject 7
0.969
0.468
0.469
0.453
0.469
0.484
0.484
0.453
0.406
9
2V4A
133
Obj NP Vs Main C sub (Version 4)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
Subject 4
While
Stevie
Participants
Subject 3
w as
smoking*
Subject 2
the
cigarette
Words
fell
Subject 1
on
his
lap
While
Stevie
w as
smoking*
the
cigarette
fell
on
his
lap
Subject 1
1.407
0.937
0.765
1.844
0.984
0.687
0.781
0.859
0.782
10.515
Subject 2
0.909
0.454
0.423
0.486
0.438
0.423
0.454
0.47
0.455
5.89
Subject 3
1.812
6.68
0.592
0.562
0.46
0.451
0.476
2.011
0.657
12.175
Subject 4
1.301
0.381
0.25
0.241
0.38
0.191
0.21
0.23
0.291
5.417
Subject 5
1.473
0.641
0.62
2.304
2.894
1.332
0.721
0.651
0.691
20.539
Subject 6
2.672
0.672
0.422
0.516
0.547
0.5
0.641
0.562
0.562
8.735
Subject 7
1.562
0.719
0.735
1.312
1.516
0.797
0.797
0.438
0.531
10.546
2V4B
Obj NP Vs Main C sub (Version 4, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
Subject 4
While
Stevie
Subject 3
w as
smoking
Participants
Subject 2
the
cigarette
Words
f ell
Subject 1
on
his
lap
While
Stevie
w as
smoking
the
cigarette
f ell
on
his
lap
Subject 1
2.11
0.656
0.5
0.453
0.531
0.687
3.375
0.687
0.609
15.828
Subject 2
1.253
0.517
0.438
0.485
0.454
0.502
0.532
0.533
0.502
12.531
Subject 3
1.016
1.794
1.283
0.618
0.449
0.524
1.185
1.154
0.496
10.418
Subject 4
0.781
0.341
0.431
0.22
0.39
0.251
0.511
0.37
0.281
5.968
Subject 5
2.714
0.861
0.852
1.071
0.691
1.642
1.883
6.329
1.132
14.821
Subject 6
2.328
0.781
0.625
2.672
0.781
0.625
0.641
0.735
0.656
11.219
Subject 7
1.406
0.797
0.625
1.187
1.625
0.922
0.64
0.578
1.078
13.485
2V5A
134
Obj NP Vs Main C sub (Version 5)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
Subject 8
Subject 7
Subject 6
Subject 5
0
dirt
was
everywhere
up*
Words
When
Participants
Subject 1
the
was
sweeping
the
princess
When
Subject 4
Subject 3
Subject 2
the
princess
w as
sw eeping
up*
w as
everyw here
Subject 1
1.5
0.5
0.375
0.312
0.328
0.312
0.36
0.25
0.25
5.297
Subject 2
1.093
0.625
0.499
0.532
0.468
0.422
0.499
the
0.438
dirt
0.515
6.937
Subject 3
13.8
0.441
0.331
0.41
0.41
0.394
0.868
0.41
0.426
4.558
Subject 4
1.128
2.096
0.568
0.559
0.619
0.519
0.579
0.729
0.548
4.581
Subject 5
1.275
0.499
0.453
0.842
0.485
0.388
0.54
0.493
0.453
Subject 6
6.234
0.691
0.739
0.947
1.003
0.928
1.043
0.792
1.003
8.879
Subject 7
4.687
0.828
0.953
0.781
1.422
2.609
0
5.906
2.875
10.281
Subject 8
2.311
0.75
0.75
0.593
0.781
0.859
0.625
0.765
0.609
7.153
4.849
2V5B
Obj NP Vs Main C sub (Version 5, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
Subject 8
Subject 7
Subject 6
Subject 5
0
Participants
Words
everywhere
was
the
Subject 1
dirt
up
sweeping
was
princess
the
When
Subject 4
Subject 3
Subject 2
When
the
princess
w as
sw eeping
up
the
dirt
w as
everyw here
Subject 1
1.578
0.453
0.328
0.422
0.344
0.515
0.312
0.265
0.359
12.203
Subject 2
1.906
0.563
0.531
0.469
0.484
0.453
0.485
0.421
0.625
24.389
Subject 3
0.879
0.345
0.282
0.439
0.329
0.407
0.439
0.298
0.392
17.141
Subject 4
0.779
0.439
0.479
0.409
0.439
0.469
0.41
0.499
0.459
8.563
Subject 5
1.051
0.421
0.389
0.341
0.357
0.373
0.365
0.373
0.373
4.738
Subject 6
1.12
0.401
0.385
0.372
0.355
0.379
0.346
0.388
0.46
7.285
Subject 7
1.671
0.64
0.516
0.531
0.609
0.734
0.718
0.578
0.734
9.496
Subject 8
1.39
0.562
0.484
0.453
0.5
0.516
0.499
0.625
0.578
4.982
2C
135
Context Obj NP Vs Main C sub (Version 3, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
Participants
was
Words
everywhere
Subject 1
dirt
the
up
sweeping
was
princess
the
when
Subject 4
Subject 3
Subject 2
w hen
the
princess
w as
sw eeping
up
the
dirt
w as
everyw here
Subject 1
0.641
0.688
0.766
0.609
0.656
0.609
0.547
0.703
0.859
11.858
Subject 2
0.52
0.491
0.421
0.44
0.441
0.451
0.49
0.501
0.451
13.79
Subject 3
0.387
0.3
0.3
0.335
0.308
0.54
0.29
0.453
0.34
14.955
Subject 4
0.18
0.188
0.22
0.188
0.235
0.5
0.245
0.364
0.324
25.284
Subject 5
0.46
0.46
0.437
0.445
0.404
0.475
0.437
0.453
0.405
Subject 6
0.349
0.381
0.349
0.549
0.34
0.372
0.412
0.372
0.388
8.793
Subject 7
0.453
0.469
0.656
0.438
0.391
0.437
0.422
0.375
0.437
11.375
4.877
3V1A
136
Dir Obj Vs Embedded Sub (Version 1)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
Subject 5
0
Subject 3
Participants
Subject 2
Subject 1
correct
not
was
answer
the
that
remembered
student
smart
The
Subject 4
Words
The
smart
student
remembered
that
the
answ er
w as
not
Subject 1
1.703
0.547
0.562
0.75
0.609
0.64
0.531
0.547
0.531
9.547
Subject 2
1.114
0.747
0.673
0.555
0.502
0.497
0.651
0.473
0.531
correct
10.646
Subject 3
1.17
0.37
0.465
0.475
0.466
0.474
0.452
0.578
0.61
6.954
Subject 4
1.843
0.691
0.635
0.676
0.669
0.69
0.81
0.594
0.635
8.102
Subject 5
1.533
0.583
0.487
0.458
0.487
0.487
0.537
0.478
0.464
6.805
Subject 6
1.395
0.641
0.563
0.5
0.498
0.475
0.673
0.563
1.608
7.84
3V1B
Dir Obj Vs Embedded Sub (Version 1, Amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
Subject 5
Subject 4
0
The
Subject 3
smart
student
Participants
Subject 2
remembered
Words
the
Subject 1
answ er
w as
not
correct
The
smart
student
remembered
the
answ er
w as
not
correct
Subject 1
1.657
0.469
0.359
0.391
0.344
0.781
0.469
0.407
17.313
Subject 2
1.268
0.931
0.701
0.668
0.484
0.581
0.525
0.716
Subject 3
1.403
0.469
0.34
0.304
0.429
0.509
0.355
0.428
9.806
Subject 4
1.521
1.148
0.717
0.603
0.572
2.451
1.1
0.659
16.861
Subject 5
1.234
0.457
0.434
0.439
0.417
0.442
0.434
0.543
16.789
Subject 6
1.7
0.524
0.544
1.061
1.346
1.123
0.947
0.644
18.501
7.871
3V2A
137
Dir Obj Vs Embedded sub (Version 2)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
Subject 10
Subject 9
Subject 8
Subject 7
Subject 6
Subject 5
5
0
Participants
Subject 4
Subject 1
insufficient
Words
was
punishment
the
that
annonced
finally
judge
The
Subject 3
Subject 2
The
judge
finally
annonced
that
the
punishment
w as
insufficient
Subject 1
1.161
0.551
0.621
0.541
0.571
0.44
0.561
0.601
7.961
Subject 2
1.938
0.466
0.264
0.248
0.264
0.434
0.465
0.341
6.916
Subject 3
1.062
0.625
0.484
0.578
0.687
0.468
0.594
1.109
6.356
Subject 4
1.903
0.51
0.461
0.591
0.851
0.5
0.521
0.961
Subject 5
0.953
0.546
0.453
0.406
0.407
0.469
0.438
0.437
6.514
Subject 6
1.125
0.562
0.61
0.485
0.485
0.453
0.672
0.687
49.905
Subject 7
1.124
0.609
0.437
0.468
0.484
0.468
1.327
0.797
Subject 8
1.156
1.364
1.293
1.453
1.074
0.866
1.092
1.043
9.191
Subject 9
1.572
0.321
0.32
0.35
0.561
0.401
0.48
0.181
12.355
Subject 10
1.182
0.521
0.5
0.501
0.421
0.42
0.421
0.431
18.005
5.668
9.415
3V2B
Dir Obj Vs Embedded sub (Version 2, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
Subject 10
Subject 9
Subject 8
Subject 7
Subject 6
Subject 5
5
Subject 4
0
Participants
was
insufficient
Words
Subject 1
punishment
the
announced
finally
judge
The
Subject 3
Subject 2
The
judge
finally
announced
the
punishment
w as
insufficient
Subject 1
2.193
1.002
0.681
1.102
0.871
0.711
0.681
23.373
Subject 2
1.336
0.384
0.261
0.445
0.199
0.43
0.215
15.861
Subject 3
1.327
0.562
0.452
0.484
0.437
0.453
0.437
15.507
Subject 4
7.789
0.952
0.901
1.111
1.072
1.121
1.051
17.543
Subject 5
1.344
0.578
0.516
0.453
0.453
0.469
0.438
13.231
Subject 6
2.422
0.515
0.546
0.625
0.532
0.5
0.813
16.937
Subject 7
2.373
0.749
0.547
0.562
0.531
0.625
0.5
16.472
Subject 8
1.908
0.908
1.49
1.477
1.101
1.66
2.38
17.478
Subject 9
2.142
0.531
0.531
0.29
0.461
0.25
0.51
15.099
Subject 10
1.072
0.631
0.791
0.731
0.551
0.701
0.541
11.366
3V3A
138
Dir Obj Vs Embedded Sub (Version 3)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
small
too
boy
Words
w arned
Participants
Subject 1
was
young
that
the
warned
prostitute
The
Instructions
Subject 4
Subject 3
Subject 2
Instructions
The
prostitute
that
the
young
boy
w as
too
small
Subject 1
1.547
0.891
0.468
0.5
0.5
0.453
0.469
0.516
0.454
0.5
6.594
Subject 2
1.412
1.041
1.102
1.151
2.143
0.591
1.322
1.062
0.481
0.45
8.653
Subject 3
38.595
1.496
0.768
0.587
0.487
0.432
0.53
0.543
0.472
0.467
11.623
Subject 4
31.393
2.004
0.662
0.532
0.464
0.489
0.455
0.428
0.446
0.441
10.437
Subject 5
14.899
2.173
0.829
0.619
0.609
0.61
0.592
0.529
0.567
0.675
Subject 6
19.135
1.598
0.616
0.505
0.438
0.426
0.392
0.44
0.448
0.459
11.25
Subject 7
10.203
1.953
0.781
0.75
0.688
0.781
0.672
0.812
0.703
0.672
10.141
6.475
3V3B
Dir Obj Vs Embedded sub (Version 3, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
Subject 4
0
The
Subject 3
prostitute
w arned
Participants
Subject 2
the
Words
young
Subject 1
boy
w as
too
small
The
prostitute
w arned
the
young
boy
w as
too
small
Subject 1
1.328
0.843
0.531
0.704
0.609
0.64
0.828
0.657
40.466
Subject 2
1.102
0.481
1.782
0.521
0.461
0.49
0.491
2.273
15.102
Subject 3
0.851
0.251
0.491
0.372
0.284
0.28
0.219
0.508
22.581
Subject 4
0.953
0.41
0.382
0.338
0.198
0.166
0.377
0.466
11.039
Subject 5
1.546
0.648
0.571
0.531
0.564
0.633
0.929
0.476
9.746
Subject 6
1.041
0.422
0.42
0.433
0.371
0.363
0.37
0.435
16.783
Subject 7
1.922
0.61
1.329
0.578
0.516
0.609
0.516
0.5
11.829
3V4A
139
Dir Obj Vs Embedded sub (Version 4)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
Participants
Subject 1
hill
the
over
was
the
way
that
woman
suggested
young
The
Subject 4
Subject 3
Subject 2
Words
The
young
w oman
suggested
that
the
w ay
w as
over
the
hill
Subject 1
1.625
0.954
0.703
0.797
1.297
0.797
0.781
4.563
4.344
0.891
63.718
Subject 2
1.081
0.548
0.61
0.455
0.47
0.455
0.532
0.517
0.532
0.579
24.64
Subject 3
1.556
0.587
0.592
0.563
0.479
0.515
0.395
0.675
0.461
0.595
Subject 4
0.922
0.26
0.32
0.211
0.4
0.221
0.39
0.371
0.29
0.19
6.56
Subject 5
1.983
0.801
1.172
1.112
0.911
1.051
0.782
0.911
0.681
0.801
18.777
Subject 6
2.657
0.657
0.5
0.5
0.516
0.625
2.25
0.75
0.625
0.734
7.656
Subject 7
1.579
0.547
0.891
0.843
0.922
1.125
0.641
0.937
0.859
1.954
12.016
9.139
3V4B
Dir Obj Vs Embedded sub (Version 4, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
Subject 4
The
young
Subject 3
w oman
suggested
Participants
Subject 2
the
Words
w ay
w as
Subject 1
over
the
hill
The
young
w oman
suggested
the
w ay
w as
over
the
hill
Subject 1
1.734
0.734
0.641
1.328
0.796
0.828
0.719
0.734
0.703
11.125
Subject 2
0.987
0.36
0.407
0.391
0.407
0.454
0.486
0.455
0.423
21.005
Subject 3
1.364
0.452
0.396
0.42
1.122
0.596
0.442
0.644
0.532
7.99
Subject 4
1.172
0.24
0.18
0.221
0.37
0.221
0.22
0.22
0.231
15.062
Subject 5
1.633
0.731
0.781
1.041
1.162
0.731
0.711
0.721
0.631
13.439
Subject 6
1.75
0.797
0.515
0.531
0.438
0.5
0.532
0.547
0.672
12.86
Subject 7
1.578
0.469
0.672
0.64
0.578
0.578
1.234
0.594
0.484
9.25
3V5A
140
Dir Obj Vs Embedded sub (Version 5)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
Subject 8
Subject 7
Subject 6
0
Subject 5
The
Subject 4
w ise
man
Subject 3
finally
believed
Participants
Subject 2
that
the
fortune
Words
Subject 1
teller
w as
talking
jiberish
The
w ise
man
finally
believed
that
the
f ortune
teller
w as
talking
jiberish
Subject 1
0.937
0.485
0.406
0.485
0.296
0.359
0.438
0.406
0.281
0.343
0.328
11.766
Subject 2
1
0.5
0.453
0.453
0.453
0.468
0.454
0.406
0.438
0.562
0.469
7.452
Subject 3
1.145
0.392
0.361
0.455
0.392
0.408
0.533
0.392
0.345
0.392
0.361
4.972
Subject 4
0.729
0.439
1.287
0.429
0.479
0.44
0.449
0.469
0.539
0.349
0.309
7.246
Subject 5
0.972
0.524
0.453
0.373
0.365
0.389
0.397
0.413
0.364
0.435
0.396
Subject 6
1.396
0.532
0.523
0.539
0.653
0.442
0.427
0.475
0.394
0.427
0.299
8.018
Subject 7
2.702
0.563
0.5
0.546
0.531
0.484
0.468
0.406
0.516
0.625
0.827
21.959
Subject 8
1.437
0.546
0.484
0.515
0.5
0.562
0.563
0.531
0.64
0.578
0.624
9.152
6.105
3V5B
Dir Obj Vs Embedded sub (Version 5, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
Subject 8
Subject 7
Subject 6
Subject 5
0
Subject 4
The
w ise
Subject 3
man
finally
believed
Participants
Subject 2
the
Words
fortune
teller
Subject 1
w as
talking
jiberish
The
w ise
man
f inally
believed
the
fortune
teller
w as
talking
jiberish
Subject 1
0.89
0.797
0.61
0.438
0.547
0.406
0.39
0.375
0.454
0.328
14.594
Subject 2
2.172
0.5
0.485
0.453
0.5
0.469
0.515
0.516
0.546
1.109
21.264
Subject 3
1.183
0.504
0.458
0.41
0.458
0.457
0.52
0.442
0.489
0.504
10.819
Subject 4
0.769
0.549
0.699
0.648
0.569
1.677
0.499
0.589
0.609
1.038
11.947
Subject 5
1.005
0.408
0.381
0.373
0.364
0.373
0.437
0.461
0.427
0.459
21.255
Subject 6
1.464
1.001
0.903
0.824
0.863
0.946
1.012
0.663
0.689
0.722
14.116
Subject 7
2.328
0.812
0.75
1.172
1.812
1.953
1.282
1.313
0.89
2.625
23.796
Subject 8
1.546
0.687
0.578
0.749
0.625
0.593
0.64
0.766
0.609
0.859
18.63
3C
141
Context Dir Obj Vs Embedded sub (Version 4, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
Subject 4
the
w ise
Subject 3
man
finally
believed
Participants
Subject 2
the
Words
fortune
teller
Subject 1
w as
talking
jiberish
the
w ise
man
finally
believed
the
fortune
teller
w as
talking
jiberish
Subject 1
0.703
0.719
0.797
0.828
1.156
1.078
2.485
1.359
0.938
0.813
26.359
Subject 2
0.454
0.329
0.438
0.486
0.391
0.407
0.313
0.407
0.47
0.36
21.507
Subject 3
0.372
0.931
0.456
0.429
0.404
0.388
0.484
0.411
0.476
0.516
21.469
Subject 4
0.24
0.28
0.251
0.29
0.29
0.241
0.31
0.231
0.22
0.32
12.739
Subject 5
0.671
0.5
0.511
0.621
0.891
0.651
0.801
0.591
0.601
0.831
70.131
Subject 6
0.578
0.594
0.484
0.453
0.437
0.406
0.469
0.469
0.484
0.516
19.875
Subject 7
1.109
0.735
0.812
0.547
0.609
0.844
0.859
0.89
0.766
1
35.75
4V1A
142
Prep Phrase Attach (Version 1)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
0
Subject 5
The
Subject 4
cook
placed
Subject 3
the
cake
that
w as
Words
Subject 2
in
the
Participants
Subject 1
oven
on
the
table
The
cook
placed
the
cake
that
w as
in
the
oven
on
the
Subject 1
1.187
0.438
0.453
0.391
0.343
0.344
0.469
0.516
0.438
0.625
0.516
0.516
7.156
Subject 2
1.164
0.429
0.443
0.485
0.5
0.509
0.54
0.493
0.456
0.683
0.62
0.5
10.047
table
Subject 3
1.019
0.356
0.372
0.469
0.394
0.452
0.434
0.435
0.363
0.372
0.426
0.428
10.005
Subject 4
2.396
0.59
0.507
0.532
0.548
0.909
0.688
0.676
0.586
0.639
1.216
0.545
10.485
Subject 5
1.311
0.458
0.378
0.399
0.378
0.388
0.386
0.426
0.399
0.45
0.419
0.41
6.261
Subject 6
3.458
0.723
0.601
0.628
0.667
1.147
0.627
0.661
0.861
0.515
0.495
0.412
28.445
4V1B
Prep Phrase Attach (Version 1, Amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
Subject 5
0
Subject 4
The
cook
the
Subject 2
cake
in
Words
The
Participants
Subject 3
placed
cake
the
oven
Subject 1
on
the
table
cook
placed
the
in
the
oven
on
the
table
Subject 1
1.734
0.5
0.484
0.454
0.5
0.5
0.453
0.921
0.531
0.562
9.765
Subject 2
1.232
0.649
0.795
0.575
0.563
0.515
0.433
0.539
0.656
0.491
35.96
Subject 3
1.083
0.409
0.431
0.459
0.61
0.466
0.435
0.497
0.474
0.49
12.854
Subject 4
1.216
0.642
0.746
0.767
0.613
0.591
0.671
0.634
0.592
1.082
14.101
Subject 5
1.065
0.429
0.485
0.544
4.699
0.59
16
0.431
0.39
0.377
6.365
Subject 6
1.77
0.585
0.905
0.667
1.067
0.81
4.105
0.961
2.721
0.859
21.762
4V2A
143
Prep phrase attach (Version 2)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
5
Subject 10
Subject 9
Subject 8
Subject 7
Subject 6
Subject 5
Subject 4
Subject 3
Subject 2
Subject 1
0
on
table
Words
the
basket
the
on
was
frog
that
put
the
princess
The
Participants
The
princess
put
the
frog
that
w as
on
the
basket
on
the
table
Subject 1
1.623
7.941
0.821
0.471
0.561
0.541
0.821
0.581
0.481
0.63
0.581
0.591
27.41
Subject 2
1.275
0.445
0.292
0.4
0.476
0.538
0.43
0.507
0.676
0.522
0.707
0.414
9.178
Subject 3
1.171
0.5
0.484
0.5
0.453
0.453
0.531
0.499
0.5
1.203
0.656
0.546
9.385
Subject 4
2.653
1.292
0.811
1.401
0.701
0.661
1.172
2.383
0.59
0.811
1.501
0.811
13.888
Subject 5
1.469
0.593
0.609
0.516
0.719
0.547
0.906
0.563
0.515
0.5
0.5
0.515
12.918
Subject 6
1.422
1.016
0.594
0.547
0.437
0.5
0.484
0.5
0.516
0.532
0.485
0.797
9.812
Subject 7
1.67
0.562
0.469
0.484
0.452
0.437
0.437
0.453
0.422
0.484
0.484
0.39
10.242
Subject 8
1.34
1.125
1.109
1.676
1.421
1.029
1.317
0.813
0.765
1.348
0.884
1.195
8.344
Subject 9
1.241
0.451
0.491
0.3
0.451
0.39
0.37
0.421
0.38
0.852
0.44
0.381
7.519
Subject 10
1.202
0.591
0.791
0.561
0.57
0.711
0.591
0.531
0.501
0.531
0.48
0.471
8.312
4V2B
Prep phrase attach (Version 2, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
Subject 10
Subject 9
Subject 8
Subject 7
Subject 6
Subject 5
Subject 4
Subject 3
5
table
the
Subject 1
on
basket
on
Words
the
Subject 2
frog
the
put
princess
Participants
The
0
The
princess
put
the
frog
on
the
basket
on
the
table
Subject 1
1.122
0.761
0.511
0.571
0.561
0.621
0.47
0.571
0.511
0.871
20.389
Subject 2
1.597
0.434
0.388
0.403
0.481
0.48
0.434
0.496
0.527
0.512
7.567
Subject 3
0.859
0.406
0.422
0.39
0.344
0.422
0.375
0.374
0.39
0.593
11.306
Subject 4
1.521
0.441
0.461
0.44
0.481
0.45
0.391
0.531
0.53
0.531
6.298
Subject 5
1.047
0.453
0.375
0.344
0.422
0.485
0.422
0.437
0.422
0.532
14.278
Subject 6
0.953
0.516
0.562
0.422
0.515
0.562
0.954
0.985
1.344
0.703
11.906
Subject 7
1.374
0.625
0.453
0.484
0.469
0.484
0.453
1.093
0.656
0.734
10.165
Subject 8
1.739
1.265
0.973
1.013
1.036
0.764
0.738
0.756
0.596
0.579
16.554
Subject 9
1.341
0.511
0.4
0.261
0.29
0.401
0.36
0.39
0.491
0.521
14.417
Subject 10
1.371
0.531
0.441
0.411
0.51
0.491
0.461
0.51
0.491
0.461
5.327
4V3A
144
Prep phrase attach (Version 3)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
The
soccer
Subject 4
player
kicked
the
Participants
Subject 3
ball
that
w as
Words
The
soccer
player
kicked
the
ball
that
Subject 1
1.281
0.547
0.844
0.672
0.516
0.437
Subject 2
1.262
0.631
0.49
0.6
0.501
0.661
Subject 3
0.858
0.207
0.276
0.255
0.268
Subject 4
1.291
0.403
0.163
0.194
0.389
Subject 5
1.036
2.076
0.57
0.502
0.417
Subject 6
0.897
0.465
0.404
0.401
Subject 7
2.578
0.5
0.578
0.578
Subject 2
on
the
penalty
spot
w as
on
0.687
0.61
0.571
0.691
0.364
0.4
0.219
0.467
0.427
0.465
0.355
0.378
0.5
0.547
Subject 1
into
the
goal
the
penalty
spot
into
the
goal
0.485
0.5
0.547
0.578
0.625
0.406
9.046
0.601
0.48
0.621
0.531
0.631
0.461
6.008
0.465
0.273
0.403
0.516
0.4
0.575
0.365
7.161
0.285
0.195
0.185
0.498
0.352
0.216
0.187
6.048
0.46
0.517
0.483
0.473
0.417
0.515
0.435
7.389
0.393
0.438
0.4
0.374
0.393
0.396
0.401
0.377
5.815
0.594
0.562
0.5
0.5
0.453
1.015
0.594
0.453
6.047
4V3B
Prep phrase attach (Version 3, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
0
Subject 5
The
Subject 4
soccer
player
kicked
Subject 3
the
ball
on
Words
Participants
Subject 2
the
penalty
spot
Subject 1
into
the
goal
The
soccer
player
kicked
the
ball
on
the
penalty
spot
into
the
Subject 1
1.281
0.625
0.563
0.562
0.578
0.562
0.625
0.532
0.563
0.985
1.875
0.656
16.53
Subject 2
1.342
1.152
1.111
1.833
0.981
0.922
1.362
2.683
0.982
1.322
1.702
0.942
43.943
goal
Subject 3
0.603
1.507
0.537
0.496
0.267
0.451
0.388
0.59
0.29
0.692
2.066
0.521
18.201
Subject 4
1.97
0.465
0.345
0.299
0.431
0.244
0.41
0.21
0.242
0.427
0.343
0.244
12.313
Subject 5
1.275
0.522
0.525
0.571
0.516
0.409
0.105
0.859
0.803
0.562
1.387
0.542
Subject 6
1.291
0.466
0.382
0.403
0.372
0.482
0.466
0.403
0.522
0.479
0.491
0.411
7.445
Subject 7
1.313
0.843
0.688
0.704
0.641
0.734
0.703
0.656
0.688
0.828
0.719
0.75
15.594
7.236
145
4V4A
Prep phrase attach (Version 4)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
Subject 4
Subject 3
Subject 2
car
to
bonnet
Words
mechanic
Subject 1
the
on
the
that
was
the
Participants
bolt
finally
attached
The
mechanic
Instructions
0
Instructions
The
finally
attached
the
bolt
that
w as
on
the
bonnet
to
the
Subject 1
36.422
2.703
1.25
0.75
0.906
0.688
1.484
0.796
2.609
0.843
0.593
3.281
1.359
0.719
10.25
Subject 2
14.739
2.569
0.862
1.786
0.705
1.159
0.862
0.925
2.678
0.736
1.3
0.878
1.488
0.69
16.572
car
Subject 3
21.204
1.315
0.728
0.787
0.696
0.474
0.818
0.719
0.623
0.69
0.634
1.242
0.659
0.505
Subject 4
7.821
1.392
0.56
0.541
0.501
0.39
0.411
0.431
0.39
0.421
0.571
0.46
0.581
0.401
5.538
Subject 5
46.187
3.865
1.933
1.152
1.211
1.122
1.282
0.931
0.862
1.011
2.213
3.285
1.502
1.092
23.033
8.809
Subject 6
32.86
11.985
2.953
1.078
1.047
0.813
0.875
0.719
1.328
0.735
0.704
0.984
1.094
1.656
15.156
Subject 7
17.844
9.328
1.922
2.109
2.828
3.281
2.109
1.859
1.453
1.093
0.875
1.235
3.094
0.859
10.484
4V4B
Prep phrase attach (Version 4, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
0
Subject 5
Subject 4
The
mechanic
f inally
Subject 3
attached
the
bolt
Participants
Subject 2
on
the
Words
bonnet
Subject 1
to
the
car
The
mechanic
f inally
attached
the
bolt
on
the
bonnet
to
the
car
Subject 1
1.64
0.829
0.562
0.468
0.421
0.484
0.562
0.563
0.547
0.547
0.625
14.828
Subject 2
0.909
0.502
0.455
0.486
0.469
0.486
0.454
0.392
0.454
1.222
0.455
19.047
Subject 3
1.194
0.531
0.579
0.476
0.449
0.443
0.443
0.853
0.52
0.561
0.494
15.263
Subject 4
0.911
0.251
0.34
0.381
0.2
0.3
0.231
0.44
0.271
0.45
0.311
8.763
Subject 5
1.833
0.931
0.841
0.771
0.65
0.902
0.641
0.621
0.831
0.701
0.801
25.617
Subject 6
2.25
0.875
0.688
0.547
0.609
0.75
0.844
0.859
0.672
0.75
1.328
15.532
Subject 7
2.875
0.5
0.562
0.61
0.5
1.579
0.687
0.656
0.937
0.579
1.593
23.188
146
4V5A
Prep phrase attach (Version 5)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
Subject 8
Subject 7
Subject 6
0
Subject 5
Subject 4
The
musician
plucked
Participants
Subject 3
the
harp
Subject 2
that
w as
in
Words
his
Subject 1
hands
The
musician
plucked
the
harp
that
w as
Subject 1
1.015
0.562
0.469
0.438
0.343
0.422
Subject 2
0.985
0.516
0.515
0.469
0.484
0.484
Subject 3
1.117
0.439
0.643
0.471
0.486
Subject 4
0.938
0.44
0.578
0.539
0.589
Subject 5
1.084
0.491
0.429
0.405
0.392
Subject 6
1.618
0.732
0.777
0.547
Subject 7
5.812
0.625
0.859
Subject 8
1.281
0.749
0.578
on
stage
in
his
hands
on
stage
0.312
0.36
0.406
0.266
0.422
13.672
0.453
0.547
0.468
0.625
0.484
6.171
0.454
0.471
0.486
0.44
0.737
0.471
5.536
0.519
0.539
0.459
0.519
0.479
0.489
6.498
0.405
0.397
0.451
0.416
0.476
0.513
0.795
1.361
0.819
0.64
0.826
0.747
2.008
7.295
0.703
0.843
0.75
0.766
0.656
0.672
1.406
0.703
10.593
0.547
0.64
0.671
0.578
0.593
0.578
0.609
0.562
7.636
5.499
4V5B
Prep phrase attach (Version 5, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
Subject 8
Subject 7
Subject 6
Subject 5
0
Subject 4
The
musician
Subject 3
plucked
the
Participants
Subject 2
harp
in
Words
his
Subject 1
hands
on
stage
The
musician
plucked
the
harp
in
his
hands
on
Subject 1
1.421
0.485
0.406
0.406
0.484
0.313
0.359
0.25
0.406
4.125
Subject 2
2.219
0.422
0.437
0.438
0.453
0.468
0.406
0.438
0.422
11.468
stage
Subject 3
1.02
0.298
0.329
0.282
0.345
0.361
0.345
0.439
0.471
5.755
Subject 4
1.048
0.469
0.441
0.45
0.23
0.39
0.38
0.51
0.43
Subject 5
0.907
0.445
0.473
0.485
0.437
0.38
0.396
0.493
0.644
4.666
Subject 6
1.003
0.516
0.504
0.45
0.427
0.458
0.458
0.465
0.5
11.515
Subject 7
2.983
0.5
0.438
0.391
0.406
0.453
0.438
0.453
0.422
9.886
Subject 8
1.124
0.546
0.703
0.484
0.531
0.546
0.531
0.563
0.624
20.239
7.193
147
4C
Context Prep phrase attach (Version 5, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
Subject 8
Subject 7
Subject 6
0
Participants
Subject 2
car
to
the
the
Words
Subject 1
bonnet
on
the
Subject 3
bolt
finally
attached
Subject 4
mechanic
The
Subject 5
The
mechanic
finally
attached
the
bolt
on
the
bonnet
to
the
car
Subject 1
0.25
0.234
0.281
0.359
1.032
0.5
0.313
0.234
0.281
0.468
0.25
14.313
Subject 2
1.984
0.5
0.453
0.468
0.672
0.578
0.593
0.469
0.828
0.578
0.563
56.419
0.58
0.298
0.314
0.36
0.22
0.439
0.298
0.33
0.267
0.423
0.235
29.404
Subject 4
0.45
0.49
0.461
0.6
0.51
0.9
0.59
0.441
0.66
1.38
0.41
13.905
Subject 5
0.443
0.564
0.578
0.469
0.436
0.469
0.405
0.445
0.453
0.429
0.492
Subject 6
0.453
0.349
0.428
0.388
0.426
0.397
0.418
0.411
0.426
0.452
0.436
7.688
Subject 7
0.891
0.781
1.421
0.671
0.61
0.718
0.625
0.641
0.999
0.703
0.531
34.283
Subject 8
0.594
0.609
0.578
0.546
0.594
0.562
0.625
0.562
0.625
0.562
0.562
15.913
Subject 3
9.032
5V1A
148
Main C Vs Comp C (Version 1)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
0
Subject 5
Subject 4
The
boat
Subject 3
that
w as
floated
dow n
Words
Subject 2
the
river
Participants
Subject 1
sank
w ithout
a
trace
The
boat
that
w as
f loated
dow n
the
river
sank
w ithout
a
trace
Subject 1
1.688
0.453
0.469
0.422
0.547
0.609
0.468
0.594
0.469
0.516
0.469
11.843
Subject 2
1.027
0.524
0.45
0.436
0.556
0.546
0.458
0.467
0.49
0.578
0.563
12.551
Subject 3
1.289
0.403
0.369
0.391
0.394
0.508
0.459
0.634
0.786
0.535
0.506
9.211
Subject 4
1.723
0.635
0.651
0.612
0.887
0.963
0.627
0.625
0.848
0.795
1.636
24.163
Subject 5
8.698
0.46
0.408
0.391
0.424
0.437
0.41
0.41
0.376
0.387
0.505
6.058
Subject 6
2.582
0.747
2.676
0.483
1.129
0.776
0.578
2.555
0.701
1.018
1.523
17.787
5V1B
Main C Vs Comp C (Version 1, Amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
Subject 5
0
Subject 4
The
boat
Subject 3
floated
dow n
Words
The
boat
floated
Subject 1
1.141
0.391
Subject 2
1.188
0.427
Subject 3
Participants
Subject 2
the
river
sank
Subject 1
w ithout
a
trace
dow n
the
river
sank
w ithout
a
trace
0.406
0.39
0.328
0.328
0.328
1.016
0.5
6.047
0.435
0.484
0.501
0.435
0.484
0.604
0.54
8.224
1.364
0.379
0.402
0.484
0.372
0.404
0.404
1.085
0.485
9.927
Subject 4
1.21
0.547
0.636
0.833
0.539
0.632
0.761
0.739
0.597
14.047
Subject 5
0.931
1.72
0.376
0.394
0.392
0.387
0.378
0.403
0.505
14.033
Subject 6
1.744
0.378
0.451
1.405
0.66
0.7
2.669
0.861
0.573
13.893
5V2A
149
Main C Vs Comp C (Version 2)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
Subject 9
Subject 10
30
25
20
Time 15
10
5
Subject 10
Subject 9
Subject 8
Subject 7
Subject 6
Subject 5
Subject 4
Subject 3
0
Participants
The
man
gave
the
dog
Subject 2
that
a
bone
Words
The
man
gave
the
dog
Subject 1
2.434
Subject 2
1.101
0.521
0.51
0.431
0.419
0.357
0.325
Subject 3
Subject 4
Subject 5
1.125
0.469
0.343
0.375
2.954
0.54
0.461
0.441
2.531
0.39
0.5
0.406
Subject 6
1.031
0.437
0.469
0.438
Subject 7
1.452
0.452
0.406
Subject 8
1.972
0.764
Subject 9
1.002
Subject 10
1.132
Subject 1
fell
on
a
bath
that
a
bone
f ell
on
a
bath
0.411
0.45
0.571
1.432
0.561
0.521
0.511
8.502
0.357
0.496
0.542
1.101
0.497
0.636
0.512
6.192
0.406
0.39
0.406
0.484
0.469
0.562
0.546
5.106
0.52
0.521
0.521
1.321
0.721
1.202
0.691
5.456
0.484
0.374
0.531
1.281
0.515
0.734
0.796
5.249
0.422
0.703
2.359
1.968
1.625
0.734
0.593
0.437
0.452
0.484
0.453
1.109
0.671
0.968
0.734
7.963
0.781
0.789
0.916
0.969
0.829
1.324
1.869
0.975
0.747
14.614
0.32
0.31
0.261
0.23
0.28
0.481
0.521
0.26
0.651
0.42
6.608
0.471
0.44
0.431
0.491
0.46
0.451
0.531
0.511
1.532
0.471
7.44
Subject 8
Subject 9
6.437
5V2B
Main C Vs Comp C (Version 2, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 10
30
25
20
Time 15
10
Subject 10
Subject 9
Subject 8
Subject 7
Subject 6
Subject 5
5
0
Subject 4
The
man
Participants
Subject 3
gave
the
Subject 2
dog
a
Words
bone
fell
Subject 1
on
a
bath
The
man
gave
the
dog
a
bone
fell
on
a
Subject 1
1.462
0.621
0.47
0.461
0.511
0.47
0.441
0.511
0.501
0.61
7.952
Subject 2
1.334
1.256
0.573
1.303
1.783
0.745
0.62
1.551
0.558
0.62
14.406
bath
Subject 3
2.139
0.516
0.625
0.499
0.5
0.469
0.468
0.656
0.531
0.64
7.387
Subject 4
2.353
0.981
1.102
0.941
1.732
1.042
0.71
0.872
1.762
0.931
6.729
Subject 5
1.297
0.453
0.437
0.438
0.453
0.485
0.437
0.437
0.515
0.719
9.435
Subject 6
1.203
0.5
0.438
0.406
0.438
0.437
0.468
1.406
1.156
0.625
9.438
Subject 7
1.374
0.546
0.39
0.499
0.437
0.484
0.437
0.453
0.499
0.5
Subject 8
1.237
1.138
1.109
1.316
2.416
0.828
1.197
4.261
1.068
0.869
7.329
Subject 9
1.161
0.621
0.551
0.42
0.25
0.49
0.401
0.42
2.974
0.42
11.564
Subject 10
1.092
0.56
0.521
0.711
0.621
0.481
0.511
0.53
1.493
0.5
6.51
4.684
5V3A
150
Main C Vs Comp C (Version 3)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
Subject 4
0
Subject 3
The
horse
that
w as
Participants
Subject 2
raced
Words
Subject 1
past
the
barn
f ell
The
horse
that
w as
raced
past
the
barn
f ell
Subject 1
0.984
0.563
0.829
0.703
0.953
1.515
0.906
1.61
30.279
Subject 2
18.627
3.064
2.924
2.444
2.764
3.965
3.505
2.935
17.184
Subject 3
1.146
0.427
0.451
0.338
0.411
0.427
0.33
0.409
9.171
Subject 4
1.337
0.498
0.267
0.387
0.194
0.191
0.722
0.205
15.246
Subject 5
1.235
0.804
0.552
0.515
0.524
0.523
0.513
0.537
4.486
Subject 6
1.201
0.484
0.442
0.426
0.448
0.441
0.457
0.515
10.083
Subject 7
1.281
0.828
0.672
0.718
0.687
0.781
0.672
0.672
7.282
5V3B
Main C Vs Comp C (Version 3, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
Subject 7
5
Subject 6
Subject 5
Subject 4
0
Subject 3
The
Participants
Subject 2
horse
raced
Words
past
Subject 1
the
barn
fell
The
horse
raced
past
the
barn
fell
Subject 1
1.031
3.297
0.593
0.547
0.532
0.641
15.734
Subject 2
0.741
0.48
0.621
0.441
0.42
0.401
10.665
Subject 3
1.21
0.299
0.234
0.42
0.276
0.259
9.962
Subject 4
1.226
0.397
0.284
0.18
0.157
0.173
Subject 5
0.899
0.394
0.489
0.538
0.461
0.513
9.631
Subject 6
1.042
0.429
0.402
0.405
0.402
0.362
10.095
Subject 7
1.062
0.516
0.5
0.516
0.453
0.547
9.969
11.22
5V4A
151
Main C Vs Comp C (Version 4)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
0
Subject 4
The
Participants
Subject 3
students
w ho
are
Subject 2
taught
in
Words
Subject 1
open
air
do
w ell
The
students
w ho
are
taught
in
open
air
do
w ell
Subject 1
3.047
0.657
0.625
0.593
0.484
0.953
0.672
0.562
0.844
13.75
Subject 2
0.924
0.47
0.392
0.391
0.454
0.47
0.486
0.956
0.501
5.921
Subject 3
0.811
0.22
0.221
0.2
0.261
Subject 4
0.811
0.23
0.22
0.221
0.2
0.2
0.261
0.24
0.18
4.958
Subject 5
2.694
0.721
0.631
0.581
0.831
2.053
2.804
1.061
0.782
48.269
Subject 6
3.531
0.828
0.875
0.609
0.641
0.562
0.625
0.906
0.797
8.062
Subject 7
1.359
0.531
0.531
0.453
0.719
0.906
1.593
0.563
1
10.922
0.23
0.2
0.24
0.18
4.958
5V4B
Main C Vs Comp C (Version 4, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
30
25
20
Time 15
10
5
Subject 7
Subject 6
Subject 5
Subject 4
0
The
Subject 3
students
taught
Participants
Subject 2
in
Words
the
Subject 1
open
air
do
w ell
The
students
taught
in
the
open
air
do
w ell
Subject 1
1.437
0.562
0.781
0.766
0.797
2.204
1.172
1.047
11.922
Subject 2
1.739
0.627
0.502
0.564
0.579
0.955
0.595
0.533
28.18
Subject 3
1.362
0.632
0.611
0.53
0.435
1.735
0.867
2.018
20.071
Subject 4
0.911
0.301
0.49
0.391
0.33
0.511
0.431
0.36
7.091
Subject 5
2.804
0.731
0.741
0.781
0.751
2.905
0.901
0.801
50.763
Subject 6
2.719
2.265
0.828
0.719
0.703
0.672
0.75
0.704
8.532
Subject 7
1.234
0.672
1.375
1.156
1.062
0.672
1.516
0.562
32.219
5V5A
152
Main C Vs Comp C (Version 5)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
5
Subject 8
Subject 7
Subject 6
Subject 5
0
Participants
the
Subject 1
Words
crashed
police
past
was
driven
car
that
The
speeding
Instructions
Subject 4
Subject 3
Subject 2
Instructions
The
speeding
car
that
w as
driven
past
the
police
crashed
Subject 1
21.766
1.297
0.703
0.469
0.359
0.438
0.578
0.406
0.375
0.359
5.219
Subject 2
19.545
2.484
0.734
0.812
0.891
0.703
1.64
0.937
1.078
1.156
Subject 3
16.702
1.561
0.583
0.568
0.489
0.441
0.489
0.505
0.504
0.442
6.798
Subject 4
5.35
4.002
3.294
2.066
1.807
0.848
0.669
0.798
0.569
0.599
10.879
Subject 5
21.372
1.22
0.402
0.388
0.349
0.333
0.395
0.324
0.405
0.331
Subject 6
14.392
2.81
1.141
0.83
1.406
0.929
0.794
1.076
0.642
1.251
9.153
Subject 7
23.092
16.812
4.624
3.437
3.61
2.64
3.968
2.39
2.984
2.516
16.577
Subject 8
17.85
4.185
1.671
1.265
1.468
1.265
1
0.874
0.703
0.812
20.441
8.515
7.249
5V5B
Main C Vs Comp C (Version 5, amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
Subject 7
Subject 8
30
25
20
Time 15
10
Subject 8
5
Subject 7
Subject 6
Subject 5
Subject 4
0
Subject 3
The
speeding
Participants
Subject 2
car
driven
past
Words
Subject 1
the
police
crashed
The
speeding
car
driven
past
the
police
crashed
Subject 1
1.032
0.468
0.407
0.36
0.468
0.688
0.328
7.234
Subject 2
0.891
0.437
0.437
0.563
0.516
0.485
0.531
Subject 3
1.207
0.376
0.423
1.035
0.502
0.471
0.486
6.304
Subject 4
0.878
0.499
0.499
0.629
0.459
0.599
0.628
11.957
Subject 5
1.059
0.524
0.46
0.354
0.381
0.357
0.389
9.368
Subject 6
1.323
0.627
1.179
0.681
0.712
0.543
0.633
10.668
Subject 7
3.75
0.75
0.703
0.656
0.719
0.531
0.75
29.639
Subject 8
1.437
0.593
0.578
0.578
0.641
0.577
0.562
12.321
6.75
5C
153
Context Main C Vs Comp C (Version 1, Amb)
Subject 1
Subject 2
Subject 3
Subject 4
Subject 5
Subject 6
30
25
20
Time 15
10
5
Subject 6
Subject 5
Subject 4
0
the
Subject 3
students
taught
Participants
Subject 2
in
the
Words
Subject 1
open
air
do
w ell
the
students
taught
in
the
open
air
do
w ell
Subject 1
0.516
0.453
0.422
0.422
0.406
0.406
0.453
0.5
6.672
Subject 2
0.476
0.425
0.404
0.485
0.46
0.484
0.5
0.445
8.526
Subject 3
0.484
0.643
0.596
0.771
0.412
0.908
0.533
0.557
Subject 4
1.02
0.746
0.666
0.687
0.637
1.109
0.917
0.949
9.561
Subject 5
0.412
0.416
0.403
0.41
0.393
0.426
0.434
0.404
21.197
Subject 6
0.62
0.532
0.5
0.58
0.5
0.883
1.004
1.116
11.279
6.799
154