Linguistics 35 (1997), 735-766
© 1997 Walter de Gruyter
The LAD goes to school:
A cautionary tale for nativists
EWA DĄBROWSKA
Abstract
Linguistics is supposed to be the study of linguistic competence. The sentences used by
linguists to illustrate their arguments are presumably structures that can be generated by
the rules that make up the native speaker's competence. Therefore, all normal native
speakers should in principle be capable of processing them. In this study, adult
respondents of various educational backgrounds were presented with a series of test
sentences based on examples drawn from recent publications in the GB framework. The
test was carefully designed to minimise the effects of extrinsic factors such as memory
limitations and lapses of attention. It was found that performance increased dramatically
with educational achievement, with the least educated respondents consistently obtaining
very low scores. An analysis of the patterns of answers given by respondents of various
educational backgrounds revealed that the least educated speakers were also the most
likely to ignore syntactic cues and rely on non-linguistic strategies in interpreting the test
sentences. Thus, the results suggest that the ability to deal with the carefully edited,
highly syntacticized structures that one encounters in publications dealing with language
is acquired in the course of formal education and is far from universal. This in turn
raises doubts about the traditional 'logical' argument for innateness.
2
1. Introduction
One of the basic tenets of generative1 grammar and its psycholinguistic offshoots is the
innateness hypothesis. Language is a system of such enormous complexity, Chomsky and
his followers argue, that it could not possibly be learned by a child who cannot conserve
volume and has problems with even the simplest of arithmetic operations (cf. Chomsky
1980a, 1980b and various other publications; Goodluck 1991; Lightfoot 1991). Yet most
three-year-olds have a fairly good grasp of the grammar of their language, and by the age
of six, the nativists maintain, their syntactic development is nearly complete. All that
remains to be learned is a few rare structures whose occurrence is, for the most part,
restricted to written language. The only way to resolve this apparent contradiction,
therefore, is to assume that the child is born with a blueprint for language; all that he or
she has to do, on this account, is to fix the values of a few parameters, and the rules of
grammar will fall out automatically.
The question that I would like to consider here is whether syntactic development
really is so rapid and whether it really is complete by the age of six. Looking at a transcript
of the speech of a typical six-year-old (1), we will find that indeed it is not very different
from informal unplanned adult discourse (2). On the other hand, it is very different from
the type of sentences that one often finds in much linguistic literature, some examples of
which are provided in (3). How do we know, then, if children do in fact command the
complex syntactic structures that proponents of Chomskyan linguistics claim they are
endowed with?
(1)
He's pushing her mom because she wouldn't hurry up. And she couldn't find the
door....the way to find the Christmas tree. She was trying to get to the other side
to get her little baby. Here--see, that's her little baby. And she was trying to get her
'cause she might get hurt. She is just a little bitty girl. (Geneshi and Dyson
1987[1984]: 153)
3
(2)
This truck comin' for us and this guy was going pretty fast and and we had passed
one car and there is no way we can get in or out and this trucker's comin' and he's
just sort of bearing down on us and honkin' his horn. He wasn't slowing down.
(Ochs 1979:67)
(3) a. John will be easy to get his wife to vouch for. (Linguistic Inquiry 1993: 161)
b. It was King Louis who I convinced that this slave might speak to. (Linguistic
Inquiry 1993: 740)
c. The militant who he arrested after learning the mayor wanted to interrogate...
(Linguistic Inquiry 1993: 737)
d. John thought that the fact that criticising himself was hard surprised Mary.
(Linguistic Inquiry 1992: 272)2
It would seem that the obvious place to look for an answer to this question is
language acquisition research. However, studies of child language have failed to provide a
definitive answer. Some naturalistic studies recorded fairly complex structures such as
those in (4) produced by children as young as three.
(4) a. It's too early for us to eat. (Limber 1973:172)
b. I think that the girl ... that's here ... doesn't .... she doesn't want me to open it.
(Limber 1973:184)
c. Me going to watch you doing your riding lesson. (Fletcher 1985:110)
Other researchers, however, were quick to point out that these occur very infrequently in
the speech of very few children, could be the result of delayed imitation, and in any case
are by no means representative. They also appeal to a host of studies which seem to
indicate that children have problems with even comparatively simple structures such as
passives (Horgan 1978; Maratsos et al. 1979) and relative clauses (Sheldon 1974, de
Villiers et al. 1979, Flynn and Lust 1980, Tavakolian 1981, Goodluck 1990), or the
resolution of pronoun reference on syntactic grounds (Shankweiler et al. 1984, Chien and
Wexler 1990, Grimshaw and Rosen 1990, Hsu et al. 1991). The other camp parries this by
pointing out that children's performance in such studies is extremely variable and is a poor
reflection of competence. Maratsos (1984), for example, argues that experimental data
4
from young children is useless as a source of evidence about their linguistic competence
because the strategies they adopt in an experimental setting mask their true knowledge.
But if language acquisition data is unreliable, how can we learn about language
acquisition?
The usual approach taken by researchers working in the generative tradition has
been to define the properties of the end product, i.e. the mature speakers' competence in
the language that the child is learning, and to hypothesize about the initial state and the
path to the final state on the basis of these. One question that no one seems to have asked,
however, is 'Does the child actually get there?' Or, to phrase it somewhat differently, can
normal adults process sentences of the kind that one encounters in Linguistic Inquiry
(which, presumably, are an illustration of the types of structures generated by the rules
that make up the adult native speaker's competence)? If they can, they should be able to
manifest this ability in an experimental setting. Of course their performance, like that of
children, is to some degree susceptible to extrinsic factors such as limitations of memory,
attention span, ability to understand the instructions, or even failure to co-operate, but the
effects of such factors can be minimised by working with motivated adult respondents and
presenting the structures in written form, which gives subjects the opportunity to
backtrack whenever necessary.
The purpose of this paper is to show that the ability to process complex syntactic
structures of the kind that one encounters in the TG/GB literature is far from universal,
and depends to a large degree on the amount of schooling that one has had; hence it
cannot simply grow out of an innate language faculty given 'mere exposure'. To test this
hypothesis, a questionnaire tapping the ability to process four syntactic structures was
developed and administered to four groups of respondents of differing educational
backgrounds. The design and results of this study are described in the following two
sections, and its implications for theories of processing are discussed in section 4.
5
2. The design of the study
2.1 Respondents
All respondents were associated with the University of Glasgow or the University of
Newcastle, either as students or as employees. The least educated group consisted of ten
cleaners and ten janitors from Glasgow, most of whom left school at sixteen and had had
no more than ten years of formal education.3 The next group consisted of ten first and
second-year undergraduates from various faculties. Since the interviews were conducted
at the end of the academic year, they had all had the benefit of at least one year of
university education in addition to twelve years at school. The third group consisted of ten
graduate students from the Arts and Social Sciences faculties, most of whom had spent at
least six years at university. Finally, the control group consisted of ten university lecturers
who had a professional interest in language, but who were not linguists--i.e. lecturers in
English literature and foreign language teachers.
The respondents probably differed not only in the amount of schooling that they
have received, but also with respect to general intelligence, willingness to engage in
intellectual activity, etc.: individuals who go to university can be assumed to be, on
average, more intelligent than those who do not; an average PhD student is presumably
both more intelligent and more motivated than the average undergraduate; and so on. This
means that any differences between the groups in the ability to processes syntactic
structures cannot be attributed to any one factor (e.g. formal education). However, if such
differences are found, they will provide strong evidence that linguistic ability is not just a
product of an innate 'language acquisition device' shared by all members of the species.
6
2.2 The questionnaire
The questionnaire (see Appendix 1) was designed to test the comprehension of four
structures. It consisted of 20 sentences: four for each of the structures and four control
sentences. The purpose of the latter was twofold: to ensure that the interviewees were
paying attention, and to prevent them from becoming discouraged should the test
sentences prove too difficult.
All test sentences were based on examples drawn from Linguistic Inquiry (see
examples (3a-d) in the previous section). They all had the syntactic structure of the
original sentence but the lexical content was modified to control for the use of nonsyntactic information in interpretation. For example, deciding who will vote for whom is
much easier in (5a) than in (5b) because an individual referred to as 'the president' is much
more likely to be voted for than someone referred to by her first name. For each structure,
there was one test sentence where general knowledge would facilitate interpretation (such
as the (a) sentence below), one where it would conflict with the syntax and hence could
interfere with the interpretation process (e.g. the (b) sentence below), and two neutral
sentences in which general knowledge would neither help nor interfere (such as (c)
below).
(5)
a. The president will be easy to get Sandy to vote for.
b. Sandy will be easy to get the president to vote for.
c. John will be hard to get his wife to vouch for. (Linguistic Inquiry 1993: 161)
The structures tested included the 'tough-movement' construction exemplified
above, sentences with complex NPs such as
(6)
John thought that the fact that criticising himself was hard surprised Mary.
7
and two variants of the parasitic gap construction, exemplified in (7) and (8). (For the sake
of convenience, these will be referred to as 'gap 1' and 'gap 2' constructions in the
subsequent discussion.)
(7)
It was King Louis who the general convinced that this slave might speak to.
(8)
The milkman who Sandy saw after learning the zoo-keeper wanted to kill got off
the bus.
The inclusion of the last two structures requires some justification. Although the
authors of the papers they were taken from appeared to believe that they were perfectly
normal English sentences, many native speakers consider them unacceptable.4 Whatever
one's view on their grammaticality/acceptability, most people will agree that one is
extremely unlikely to encounter structures of this type outside of Linguistic Inquiry and
other journals of a similar theoretical orientation. They have been included in this study
because they shed light on two subsidiary issues which will be taken up later:
(1) Are the sentences one finds in Linguistic Inquiry English, or are they GB-speak?
(2) When processing these structures, do people rely on normal processing strategies or
do they develop ad hoc strategies for dealing with them?
Each test sentence was accompanied by two to four questions. The questions were
all very simple, and to make scoring easier, they were as far as possible closed questions
(i.e. there were only 2-3 possible answers). For example, the questions corresponding to
Q2,5 Sandy will be easy to get the president to vote for, were:
(9) a. Who will vote?
b. For whom will this person vote?
c. What will be easy?
d. Who will find it easy to do this?
Since it was found in a pilot study that respondents often relied on simple testtaking strategies in answering the questions, care was taken to minimise their effect on the
8
final results. For example, given the test sentence Q16, It's the ticket conductor who Jim
managed to convince that the red-nosed man was looking for, and the questions in (10),
(10) a. Who convinced someone of something? [Jim]
b. Whom did this person convince of something? [the conductor]
c. Who was looking for someone? [the red-nosed man]
many respondents gave the correct answer to the last question without thinking, simply
naming the participant who had not been mentioned in any of the previous answers. Thus,
any respondent who correctly identified the subject of convince (which is not very
difficult) would have a 50-50 chance of answering the other two questions correctly
simply by guessing, and the effects of this simple test-taking strategy would obscure the
effects of syntactic processing. In order to avoid the data being distorted in this way, two
of the questions corresponding to test sentences with three participants were about the
same participant (for example, the last question for the test-sentence cited above was Who
was the red-nosed man looking for?). As a result, the respondents who guessed would
give an incorrect answer to at least one of the questions.
2.3 Method
To prevent subjects (particularly the less educated ones) from getting nervous, the
interview was very informal. All respondents were warned that some of the sentences that
they would be presented with would be rather strange and difficult to make sense of, and
told that they should not worry if they could not answer some of the questions, but they
were urged to do the best they could. Since the purpose of the study was to obtain the
best possible reflection of competence, it was important to eliminate performance factors
such as memory limitations, impaired perception, and lapses of attention. Therefore,
subjects were given all the test sentences in writing, which allowed them to backtrack
whenever necessary and to process the structures at their own pace.
9
On the other hand, the questionnaire was supposed to test language
comprehension, not literacy. If the test sentences were presented in written form only, it
could be argued that the results obtained by the least educated speakers were lower
because of poor reading skills rather than an imperfect knowledge of syntax: in other
words, that they wasted so much effort on decoding the written symbols themselves that
they had very little processing capacity left for higher-level processing. To minimise the
effect of reading skills on performance, all sentences were read out to the respondents
during the interview, and repeated as many times as they wished.6
However, the version of the questionnaire that the respondents received did not
contain the questions, but only the test sentences themselves. This was done for two
reasons. First, it made it natural for the interviewer to read the test sentences out loud to
the respondents, since it was a way of indicating which sentence the following questions
were about (during a pilot study, some subjects felt insulted when the sentences were read
out to them, and insisted on reading them themselves). Secondly, it minimised the
probability of the respondents' previous responses affecting their answers to subsequent
questions. In the pilot study, subjects frequently gave contradictory answers, and then
corrected themselves when they realised this, which was much easier when the questions
are printed alongside the test sentences on the questionnaire that the subjects received. For
example, in response to (10) above, many respondents said 'Jim' in response to question
(a) and 'Jim' again when asked (b), but on re-reading the questions, they realised that Jim
cannot be both the subject and the object of convince, and changed their answers. Not
having all the questions simultaneously available reduced the effects of such 'post-editing',
as respondents had to go back to the test sentence after each question to retrieve the
information.7
For the control group (i.e. the lecturers) the last step in the procedure was to
collect grammaticality/acceptability judgements about the test sentences. Before asking for
these, it was explained to them that 'grammatical' was not to be understood in the
10
grammar-textbook sense (i.e. that they should ignore 'mistakes' such as dangling
prepositions and the use of who instead of whom), and that their task was to decide
whether a sentence was 'English' (albeit rather inelegant in some cases), 'borderline', or
'word salad'.8
2.4 Scoring
The scoring system in this study had to meet two requirements: it should provide as
faithful a reflection of the respondents' linguistic knowledge as possible--i.e. it should
minimiae the effects of guessing; and it should yield comparable results for all groups. The
implications of the two requirements are discussed below.
2.4.1 Minimising the effects of guessing
If 'who did X' type questions are asked about a test sentence with three participants, the
probability of the subject's giving the correct response by guessing is 1/3. If there are three
questions, mere guessing will, on average, yield one correct answer. To minimise the
effects of guessing, points were awarded only for performance above chance level, i.e.
three correct answers out of three gave the maximum number of points (4), two correct
answers yielded 2 points, and one correct answer was not awarded any points at all. When
there were only two questions, two correct answers yielded 4 points and one correct
answer 2 points. The tough-movement sentences had four questions each (cf. (9)), one of
which (d) was more difficult than the other three, since the answer was someone who was
not mentioned in the sentence.9 Because of this difference, the following scheme was
adopted: 4 points if all answers were correct; 3 points if all answers except (d) were
correct; 2 points if the respondent made a mistake on a question other than (d); 0 points
for more than one mistake.
11
2.4.2 Obtaining comparable results for all groups
Another difficulty with scoring the answers was that the various groups tended to give
different types of answers. Most of the educated speakers refused to answer at least one of
the questions and claimed that the sentence didn't make sense and no answer was possible.
The less educated speakers, on the other hand, were more likely to guess when they were
unable to answer a question. This difference in response patterns is due to social factors:
academics, particularly those with a professional interest in language, feel that they 'own'
language, and are in a position to decide what is and what is not acceptable. Cleaners and
janitors, on the other hand, when faced with a sentence that they could not make sense of,
tended to feel that there was something wrong with them, not the sentence (e.g. they were
not intelligent enough, or did not have enough education). The problem was how to score
responses such as 'This doesn't make sense. You can't answer this question on the basis of
the information in the sentence'. If a number of well-educated speakers refuse to answer
the same questions and moreover most judge the corresponding sentences as
ungrammatical, one can hardly score their answers as incorrect. On the other hand,
scoring them as correct would inflate their performance in comparison to that of the less
educated respondents, who may also have refused to give an answer had they had more
self-confidence.10
To ensure that the differences in scores between the groups were not an artefact of
the scoring system, two complementary scoring methods were used. The first of these ('Ascores') was the method described above, which penalised respondents who refused to
answer (points were awarded for each correct answer; failure to answer was scored as
incorrect). The second scoring method ('B-scores') penalised respondents who guessed.
These scores were obtained by calculating the number of sentences for which the subject
gave at least one incorrect answer and subtracting it from 16, i.e. the total number of test
12
sentences.11 Thus, under this method, points were awarded for any set of questions for
which the respondent did not give any incorrect answers, and there were no penalties for
failing to answer.
3. Results and discussion
3.1 Results
The results of the interviews are presented in Table 1 (A-Scores, i.e. scores based on the
number of correct answers) and Table 2 (B-Scores, or scores based on the number of
sentences without errors).
Structure
Cleaners
Janitors
Undergrads Graduates
Lecturers Pts available
complex NP
46
22
60
106
144
160
'tough'
65
73
98
126
153
160
gap 1
70
74
100
130
130
160
gap 2
86
104
118
140
144
160
Total
267
273
376
502
571
640
Table 1: A-Scores
Structure
Cleaners
Janitors
Undergrads Graduates
Lecturers Pts available
complex NP
6
2
9
22
34
40
'tough'
0
0
10
19
37
40
gap 1
13
15
25
27
36
40
gap 2
15
20
29
33
37
40
Total
34
37
73
101
144
160
13
Table 2: B-Scores
The overall scores for each of the groups are presented graphically in Figure 1.
600
500
400
300
A-scores (points for
correct answers)
200
B-scores (number or
sentences without
mistakes) x 4
100
0
Cleaners
Janitors
U ndergraduates
Graduates
Lecturers
Figure 1: Total scores
Two facts are immediately apparent from the data presented in Figure 1.12 First,
there is a very clear progression in the scores, with the least educated respondents
obtaining the lowest scores and the best educated respondents the highest. Thus, the
hypothesis put forward in section 1 is confirmed. Secondly, the two scoring methods did
not produce significantly different results: the two curves have very similar shapes. It is
true that the differences between the groups are more pronounced for B-scores, but this
was to be anticipated, given the fact that the less educated speakers made more mistakes,
and hence a significant proportion of their A-score points came from test items for which
they got only partial credit. Since under method B, points were awarded only when the
respondent gave correct answers to all of the questions that went with a particular test
item, B-scores for these respondents are lower. As we move along the X axis, the gap
14
between the two scores becomes narrower. For the best-educated speakers, who obtained
very high scores, the difference between the two scores is negligible.
160
U nskilled
140
U ndergraduates
Graduates
120
Lecturers
100
80
60
40
20
0
complex
NP
tough
gap 1
gap 2
controls
Figure 2: A-Scores on each sentence type
Figures 2 and 3 present the scores for each sentence type. (The cleaners' and the
janitors' scores have been collapsed since they are not significantly different.) Again, we
have a very clear progression according to educational attainment. (It is worth noting,
however, that the rise is steeper for some sentence types than for others. This point will be
taken up later.)
15
40
Cleaners
35
U ndergraduates
30
Graduates
Lecturers
25
20
15
10
5
0
complex
NP
tough
gap 1
gap 2
control
Figure 3: B-Scores on each sentence type
3.2 Discussion
The crucial question posed at the beginning of this study was: do all adults possess the
capacity to process complex syntactic structures of the type that one encounters in the
TG/GB literature? The data presented above appears to suggest that the answer is 'no'.
Many respondents in this study were presented with the test sentences under ideal
conditions, and yet were unable to answer simple comprehension questions about them.
The most obvious reason for this inability is that they the lacked linguistic knowledge (i.e.
competence) necessary to deal with the structures exemplified in the test sentences.
One could, of course, reject this conclusion and claim that the competence was
there, but for some reason the respondents were unable to access it. However, an
argument of this kind remains vacuous unless supported with positive evidence for the
existence of such 'invisible' competence and a plausible explanation of why the respondents
were unable to access it. (It should be remembered that performance was nearly perfect on
the control sentences, which means that the poor results on the test items cannot be
attributed to such extrinsic factors as lack of familiarity with the testing situation.)13
16
If it is true that the less educated respondents in this study lacked the linguistic
competence to deal with the test structures, then it cannot be true that all normal humans
are born with a language faculty (universal grammar, language acquisition device, etc.),
which, upon exposure to language data, inevitably leads to the development of
competence. This in turn raises serious doubts about attempts to reconstruct the
properties of the initial state of the language acquisition device on the basis of a 'native
speaker competence' supposedly shared by all native speakers of the language.14
In the remainder of this section, I will provide further evidence which indicates that
at least some subjects in this study lacked the competence to process the sentences that
they were confronted with and relied on other cognitive abilities to answer the test
questions.
Normal linguistic processing is automatic, effortless, and extremely fast. The test
sentences used in this study turned out to be very difficult for all subjects, even the
lecturers, and required a great deal of effort to process; many respondents, particularly in
the least educated group, were visibly tired at the end of the interview. Respondents in all
groups took a long time to answer the questions (sometimes as much as 2-3 minutes per
sentence), which suggests that they did not have access to ready-made routines for
processing the structures, and, as we will see shortly, many subjects tended to rely on nonlinguistic strategies. Finally, many respondents, particularly the less educated ones, gave
inconsistent answers, which provides further evidence that they did not have a unified
representation of the test sentences and relied on ad-hoc strategies to answer the
questions.
For example, consider the following responses to Q15, That little boy will be very
difficult to persuade his teacher to criticise. (In the examples below, CLE stands for
'cleaner' and JAN for 'janitor'.)
(11)
Q:
Who might criticise someone?
CLE3: The little boy.
17
Q:
Who might be criticised?
CLE3: The teacher.
Q:
What will be very difficult?
CLE3: To persuade the teacher to criticise the boy.
Q:
Who will find it difficult to do this?
CLE3: The little boy.
Two other respondents' answers to Q14, The mayor who Julie warned after learning the
ex-prisoner wanted to interrogate managed to get away, can be used to illustrate the same
point. The questions for Q14 were
(12) a. Who wanted to interrogate someone?
b. Who was supposed to have been interrogated?
c. Who managed to get away?
JAN3 answered 'the mayor' on all three questions, while JAN9 said 'the prisoner'.
3.2.1 Non-syntactic strategies
Respondents differed not only with regard to the number of correct answers that they
gave, but also in the types of mistakes that they made; these provide some clues about
how they processed the test sentences. The less educated speakers tended to rely more on
simple pragmatic strategies and real-world knowledge and less on syntactic cues. For
example, they tended to assume that all their answers must make reference to an individual
named in the test sentence, and hence nearly all gave incorrect answers to the question
about who will find it tough etc. to do something in the tough-movement sentences.
Nearly all the cleaners and janitors said that the person who will find it difficult to do
something in Q15 is either the little boy or the teacher, even if this resulted in a
contradiction (see above). This type of response was also quite common among
undergraduates, and correlates most clearly with intellectual sophistication. It is this
18
question which is responsible for the B-score of 0 for this construction obtained by the
least educated group, where 18 out of 20 respondents invariably named a participant
referred to in the sentence (even though many produced answers such as 'someone',
'whoever', 'some other person' on other questions). The two respondents who gave correct
answers to one of these questions made mistakes elsewhere on the same test sentence, and
neither of the two consistently said 'someone else' on all questions of this type. (It is worth
pointing out that answers such as 'I don't know', 'You can't tell', 'It doesn't tell you' were
scored as correct for this question.)15
Secondly, many respondents, particularly the less educated ones, tended to assume
that all individuals mentioned in the sentence must be named in the answers; as a result,
they often named the hitherto unnamed person in response to the last question in each
series. The following responses to the questions accompanying Q16, It's the ticket
conductor who Jim managed to convince that the red-nosed man was looking for, will
illustrate the point.
(13)
Q:
Who was the red-nosed man looking for?
JAN4: The ticket conductor.
Q:
Who convinced someone of something?
JAN4: Jim.
Q:
Who was convinced of something?
JAN4: The red-nosed man.
Thirdly, when syntactic signals conflicted with expectations based on relative
plausibility, the less educated speakers often relied on the latter. Thus, 35% of the
unskilled workers group said it was Sandy who will vote for the president in Q2, Sandy
will be easy to get the president to vote for, and 15% couldn't answer the questions about
who will vote for whom at all. For Q14, The mayor who Julie warned after learning the
ex-prisoner wanted to interrogate managed to get away, only 30% said (correctly) that it
was the mayor who was supposed to have been interrogated. And none of the least
19
educated group were able to say what the criminal realised or what would not be
overlooked by the investigators in Q18, The criminal realised that the fact that leaving
one's fingerprints on the revolver was dangerous would not be overlooked by the
investigators. (The most common answer for both questions was 'that he had left his
fingerprints on the revolver'.)
The last example can also be used to illustrate another simple strategy often
resorted to: many respondents, having identified the major constituents in the sentence,
assumed that whatever was close together belonged together syntactically--i.e. they relied
on simple proximity relations rather than on hierarchical structure in interpreting the
sentence. In the example just cited, most respondents claimed that the clause immediately
following realised (leaving one's fingerprints on the revolver) was its object. This was a
very common mistake even in examples where the semantic content did not strongly bias
the respondent towards the incorrect response. For example, the following answers were
obtained from the least educated speakers for Q9, John remembered that the fact that
criticising himself was hard surprised Arthur:16
(14)
Q: What did John remember?
A:
(15)
criticising himself/Arthur's criticising himself 40%
that criticising himself was hard
45%
that Arthur was surprised (correct)
15%
Q: What surprised Arthur?
A:
that criticising himself was hard (correct)
20%
that John remembered
35%
that John was criticising himself
45%
This is an extremely interesting type of mistake, and its significance will be discussed is
some detail in the next section.
3.2.2 Two puzzles
20
I would like to conclude this section with two curious facts which emerge from the data
presented above. One could claim that the low scores obtained by many respondents in
this study can be attributed to the fact that most of the test sentences were very unusual
structures of borderline grammaticality, if they were grammatical at all. Consequently,
critics might argue, the fact that people failed to give 'correct' answers does not show any
deficiency in linguistic competence, but merely indicates that the sentences that the
interviewees were confronted with do not belong to the set of sentences that make up the
English language. This may well be a valid point (though, if so, it does say something
about the validity of the research in the journal from which the test sentences were
drawn!), but the path from premise to conclusion is not as straightforward as it might at
first seem. This is because, for the sentences in this study, grammaticality is a poor
predictor of performance, particularly for the less educated speakers, whose lowest scores
were on the only test structure that was uncontroversially grammatical (the sentences with
the complex NP the fact that S) and who obtained much higher scores on the parasitic gap
constructions (cf. the data in Table 1 above).
This effect is largely due to the fact that the complex-NP and the parasitic-gap
tasks required the extraction of different types of information from the test sentence. In
the case of the former, respondents had to establish the relationships between the main
clause, a clause embedded inside it, and another clause embedded inside the subordinate
clause. For the parasitic gap sentences, on the other hand, all they had to do was to
correctly identify the subject or object of a verb. The correct answer was always one of
the three participants mentioned in the sentence, and some of the answers could be arrived
at by means of very simple syntactic strategies and the process of elimination. For
example, it does not take much knowledge of syntax to identify the subject of convince
(i.e. answer the question 'Who convinced someone?') in Q3, It was King Louis who the
general convinced that this slave might speak to. Once the respondent has established that
21
the person who did the convincing was the general, he or she will know that 'the general'
cannot be the answer to the next question, 'Who was convinced of something?'. This
leaves King Louis and the slave--and the respondent has a 50-50 chance of giving the right
answer just by guessing. One can easily get 6-8 points in this way; the average score on
these sentences in the unskilled workers group was 8.4. Nevertheless, it is still striking that
the least educated speakers' scores on the parasitic gap sentences should be so much
higher than those obtained for the other two sentence types.
Another curious fact can be observed by looking at the curves traced by the top
ends of the bars in Figures 2 and 3 above, which rise much more steeply for the first two
constructions than for the last two. In other words, although the more educated speakers
invariably obtained higher scores on all construction types, the differences between the
groups were much less pronounced on the parasitic gap sentences. This, I would like to
suggest, is due to differences in the way these groups habitually process language. The
least educated groups, we have seen above, tend to rely on simple processing strategies,
such as associating constituents which occur near each other in a sentence. These
strategies work very well for the type of language they encounter in their daily lives:
unplanned informal discourse. When faced with more syntacticized language (e.g. a
newspaper article or a formal speech), they can use redundancies and real-world
knowledge to help them work out the meaning. These strategies, however, do not work
very well on the test sentences, which have been designed to test the use of syntactic
information.
The more educated groups, on the other hand, tend to rely more on syntactic
strategies. This is not to say that individuals who have had less schooling do not use
syntax, or that highly educated speakers do not use non-syntactic strategies. Rather, what
I am suggesting is that the more educated the speaker, the more he or she is likely to make
better use of syntactic information, this being a skill that is acquired in reading and writing
highly syntacticized texts such as academic textbooks, journal articles, legal documents,
22
etc. Unplanned informal discourse is very simple syntactically,17 so individuals who are
exposed primarily to this type of text either do not develop or do not fully automate the
routines necessary for processing e.g. multiply embedded clauses. These, on the other
hand, are the daily bread and butter of academics. Therefore, on encountering a sentence
such as Q12, Paul noticed that the fact that the room was tidy surprised Shona, a
university lecturer will simply apply his or her knowledge; an individual who has not had
the benefit of a long formal education, on the other hand, may simply not possess the
relevant cognitive routines. When faced with a sentence with a parasitic gap, on the other
hand, our lecturer is in more or less the same position as the cleaner: since these have been
invented by linguists working in the Chomskyan framework, the lay person does not have
the cognitive routines for processing them, and has to extrapolate from his or her
knowledge rather than merely apply it. Hence, processing requires more effort and is more
time-consuming, and, because all respondents are in the same position, differences
between groups are much less pronounced.
4. Some implications for language processing
We now turn to the implications of the study described here for theories of language
processing. This may at first seem presumptuous: after all, language processing is
prodigiously fast, the relevant time spans being measured in milliseconds, whereas the
responses recorded in this study were produced after considerable delay, in some instances
as much as 2-3 minutes after subjects had heard the test sentence. This being the case,
how can one disentangle the output of the language processing system from non-linguistic
inferences? This is a perfectly valid objection; however, the data collected in this study
does warrant at least two conclusions about on-line processing.
First, the very fact that respondents often took so long to process the test
sentences suggests that they do not have the cognitive routines required for processing
23
them--otherwise, why wouldn't they use them? This in turn suggests that the type of
competence that work in the TG/GB paradigm is supposed to uncover has little to do with
the competence that people bring to bear in normal language processing.
Secondly, the evidence from this study creates a serious problem for modularist
theories of language processing (Frazier 1978, Forster 1979, Fodor 1983). We have seen
in section 3.2.1 that many respondents, particularly the less educated ones, tended to rely
on semantic and pragmatic strategies rather than on syntax, especially when syntactic and
semantic/pragmatic information pointed in opposite directions. For example, many
respondents said that it was Sandy who voted in Q2, the mayor who wanted to interrogate
the ex-prisoner in Q14, and that what the criminal realised in Q18 was that he had left his
fingerprints on the revolver.
Q2
Sandy will be easy to get the president to vote for.
Q14
The mayor who Julie warned after learning the ex-prisoner wanted to interrogate
managed to get away.
Q18
The criminal realised that the fact that leaving one's fingerprints on the revolver
was dangerous would not be overlooked by the investigators.
The modularity thesis maintains that syntactic processing is impenetrable to extralinguistic
belief and that other cognitive systems involved in language processing can only operate
on the input provided by the language module. If this were the case, people could not
arrive at interpretations which were inconsistent with the syntax. In other words, the
modularity thesis predicts that some of the responses recorded in this study should not
occur.
It could be argued, of course, that the situations described by these sentences are
so wildly improbable that on hearing them the listener mentally 'corrects' the speaker,
assuming that he could not have meant what he said--that is to say, that in the case of
some extremely unlikely utterances, plausibility judgements can override the interpretation
provided by the language module. However, there are several problems with this
24
argument. First, it is not immediately obvious why the less educated respondents should be
more inclined to 'correct' the sentence than the control group of university lecturers. (It is
worth noting that all the respondents were aware that the sentences were 'tricky' and some
regarded them as genuinely ambiguous; moreover, many gave pragmatically implausible
answers to other questions--cf. footnote 15.) Furthermore, presumably none of the
respondents would have interpreted the sentence below as meaning that Sandy voted for
the president, that the ex-prisoner wanted to interrogate the mayor, and that the criminal
realised that he had left his fingerprints on the revolver, respectively.
(16)
The president voted for Sandy.
(17)
The mayor wanted to interrogate the ex-prisoner.
(18)
The criminal realised that it was dangerous to leave one's fingerprints on the
revolver.
Most significantly of all, the above account cannot explain why people sometimes abandon
a perfectly plausible and syntactically appropriate interpretation in favour of a syntactically
inappropriate one. For example, when presented with Q12, Paul noticed that the fact that
the room was tidy surprised Shona, 80% of the janitors said that what Paul noticed was
that the room was tidy, and 40% said that what surprised Shona was that Paul noticed. It
is quite clear that was is happening here is that even though the syntactic relationships are
clearly marked (that the fact that), some of the respondents disregard important syntactic
cues, and jump to conclusions based on semantics. This is not to say that they do not make
use of syntactic information; however, they might rely on non-syntactic strategies when
the sentence is difficult, and the less schooling they have had, the more they are likely to
do so. This does not fit in very well with the claim that syntactic processing is independent
of other cognitive processes and that the neural architecture that supports it is innate. On
the contrary, it seems to indicate that syntax is learned (to a large degree at school and
university) and that when processing a sentence, people make use of whatever information
is most readily available.
25
It seems that the best way of accounting for the available data is the following: The
most deeply entrenched18 (i.e. the most highly automated) cognitive routines are the
fastest. Since basic syntactic routines are automated at a very early age, and are among the
most deeply entrenched, when simple sentences are processed, syntactic routines will
nearly always outrun pragmatic inferencing and other 'higher level' processes. On the other
hand, the routines for processing complex structures are acquired later, and may never
become (fully) automated. If a routine is only partially automated, or not automated at all,
processing takes time, and the language user may very well have time to retrieve the
relevant semantic/pragmatic information and make the necessary inferences before the
syntactic analysis is complete, in which case he/she will skip certain stages of syntactic
analysis (or perhaps use semantic/pragmatic information to guide parsing).19 Given the fact
that normal text is highly redundant, this is a very sensible strategy. Now obviously the
hearer cannot rely on semantic and pragmatic information only: some preliminary syntactic
analysis will have to be undertaken if he or she is to make any use of other sources of
information. But this is only to say that an open, interactive organisation is more efficient
for processing complex structures.
Let us backtrack a little. Fodor's best argument for modularity is speed: an
informationally encapsulated system does not need to consider all the information that
might in principle be relevant in processing the input, and hence can produce a response
much faster than a system which has to consider everything an individual knows before
responding. Thus, according to Fodor, a modular cognitive architecture is more efficient
than an interactive one. He also points out, when arguing for the innateness of input
systems, that the relative facilitation of information flow (by hardwiring certain
connections) amounts to the same thing as relative encapsulation. Now acquired
automation also results in the facilitation of information flow. It is not surprising,
therefore, that we may get modularity-like effects when basic syntactic structures are
processed: since the necessary syntactic routines are highly automated, and hence very
26
fast, they will often outrun other modes of processing. This does not mean that
semantic/pragmatic processing is not taking place in parallel with syntactic analysis: all it
means is that it sometimes lags behind syntactic processes, and, when it does, it cannot
influence parsing.20 In the case of more complicated sentences, however, the other modes
of processing will often outrun syntax. What this means is that from the point of view of
efficiency, it makes sense to make syntactic information available to other processes as
early as possible (i.e. even before syntactic analysis is complete) and to feed the results of
semantic and pragmatic processing back into syntax as soon as they become available.
This, of course, amounts to saying that an interactive system will perform more efficiently
than a modular one.
5. Concluding remarks
The data presented in this paper shows quite clearly that the ability to process complex
syntactic structures, or at least the complex structures that are the stuff of the arguments
in the TG/GB literature, is not universal, and hence cannot be the inevitable result of an
innate language faculty.21 On the contrary, this ability is very strongly correlated with the
amount of schooling that an individual has received, which suggests that it is a product of
literacy, not genes. Indeed, the cognitive routines for processing complex syntactic
structures are of little use when dealing with normal everyday language, which tends to be
very simple syntactically (Givón 1979, Ochs 1979, Miller 1994, Brazil 1995; but see
footnote 17). It is reasonable to assume, then, that they are acquired and become
automated when processing written text. The sort of competence that is necessary for
making sense of the types of sentences that linguists like to study is a schooled language
competence, typical of a highly literate culture and certainly not biologically specified. (It
is worth noting that even the least educated respondents in this study have had at least
eight years of schooling, and hence are not very good specimens of an unschooled
27
linguistic competence. The latter are extremely difficult to come by in a developed society
with compulsory education, so we can only speculate about how a completely illiterate
individual would have done on a test of the kind that was described here. Perhaps it is not
a coincidence, however, that much of the experimental evidence for modularity (e.g.
Rayner et al. 1983, Ferreira and Clifton 1986) has come from studies of eye-movements in
reading, and that the subjects in these studies are university students.)
In the remaining parts of this paper, I would like to do two things. First, I would
like to look at a possible objection to the conclusions proposed in the preceding sections
and discuss its validity. This will provide part of the answer to the question that I will
address in the final section, namely, why has the data of the kind collected in this study
been ignored for so long?
5.1 The performance/competence terminology game
The generativists have evolved several strategies for parrying arguments based on any
known facts about how people actually use language. Of these their favourite is playing
what George Lakoff (1973) calls the 'performance/competence terminology game'. The
strategy allows players to dismiss anything that does not fit the proposed linguistic theory
(e.g. the fact that some speakers are unable to process structures which are supposed to be
representative of the output of their internal grammars) by saying that it belongs to the
domain of performance, and hence is of little relevance to linguistics, which is the study of
competence. Other convenient dichotomies which can be used to argue away any
recalcitrant data are acquisition/development (which enables the researcher to say that a
child who obviously cannot deal with a structure that he or she is supposed to have
mastered has the requisite competence, but is unable to use it) and core/periphery (a lastditch defence, whereby anything that doesn't fit can be claimed to fall outside the scope of
'core grammar', and hence be of little interest).
28
My response, of course, is that the competence/performance distinction is useful to
the extent that it allows the linguist to abstract away from linguistically irrelevant factors
such as lapses of attention, memory limitations, and speech errors. However, it cannot be
used to argue away any results that contradict the proposed solutions. If highly motivated
respondents cannot correctly interpret a sentence under ideal conditions, i.e. when it is
presented to them in both spoken and written form and they have all the time they wish to
process it, then they lack the competence to deal with it. Turning over to Jerry Fodor,
"...some linguists have recently claimed that no data except certain specified kinds
of facts about the intuitions of native speakers could, in principle, be relevant to
the (dis)confirmation of grammatical theories. Experimental observations from
psycholinguistics are thus ruled irrelevant a priori. However, this sort of
methodology seems a lot like special pleading: you tend to get it precisely when
cherished theories are in trouble from prima facie disconfirming data." (1983:106)
A more serious objection to the conclusions proposed here could be made by
invoking social factors. The low-scoring respondents in this study, it could be argued, are
likely to have had rather negative experiences with the school system, and could have been
educated out of believing that they can deal with complicated structures of the type that
were used in this study; or they may simply belong to a population which does not
perform well on tests of any kind. I concede that both of these factors may have had some
effect on the performance of the least-educated group, but there are good reasons to
believe that the magnitude of the effect could not have been very large. First, 65% of the
cleaners and janitors gave correct answers for all control sentences; their mean score was
14.6, which is not radically different from that obtained by the other three groups in the
study (16). If lack of self-confidence and unfamiliarity with the testing situation depress
performance, they presumably depress it across the board, and not only on some sentence
types. Secondly, one could hardly propose explanations of this kind to account for the
performance of undergraduate students, whose scores, though significantly higher than
those of the least-educated group, were also disturbingly low.
29
5.2 Linguistics and ideology
Perhaps the most astonishing fact about the conclusions outlined above is that they are so
obvious. Few non-linguists would be surprised if they were told that the ability to interpret
complex sentences differs from person to person and is intimately tied up with educational
attainment. Educators have certainly known for a long time that command of complex
syntactic structures is acquired slowly and with considerable effort, and is highly
dependent on individual ability and motivation. Teachers have taught syntax for many
years, and there is a sizeable body of literature on how to do it (see e.g. the papers cited in
Hunt 1977). Why, then, has something that seems so patently obvious been ignored by
linguists for such a long time?
Part of the answer lies in the fact that most of these discussions are to be found in
books on how to teach reading and writing, whereas linguistics is supposed to be primarily
about spoken language. This is one of the first things that an undergraduate beginning
linguistics is told, and it is a myth. Generative linguists in particular pay only lip service to
this principle, and tend to focus on highly syntacticized structures which are practically
non-existent in spoken discourse: the use of who rather than whom in the examples, and
throwing in the occasional colloquial expression are merely token gestures. However, the
official line is that the carefully edited examples that one finds in linguistic journals are a
better reflection of the set of rules that we use when we speak than are any real texts,
including actual conversations.
But there is another reason, even more deeply ideological. Many people believe
that there is a connection between the language that one speaks and one's intelligence
and/or personal worth. Folk-linguistic attitudes are frequently expressed in moral or quasimoral terms:
bad grammar, slovenly speech, sloppy pronunciation, foul language
30
(Gordon 1981), and up until fairly recently, value judgements of this kind pervaded even
scientific and official government publications, as the following excerpts will illustrate:
"Our estimation of the language of culturally deprived children agrees, however,
with that of Bernstein, who maintains that this language is not merely an
underdeveloped version of standard English, but is a basically non-logical mode of
expressive behaviour which lacks the formal properties necessary for the
organisation of thought. From this point of view, the goal of language training for
the culturally deprived could be seen as not that of improving the child's language
but rather that of teaching him a different language which would hopefully replace
the first one, at least in school settings." (Bereiter et al, quoted in Gordon 1981:56)
"...the first and chief duty of the Elementary School is to give its pupils speech--to
make them articulate and civilised human beings, able to communicate themselves
in speech and writing, and able to receive the communication of others. It must be
remembered that children, until they can readily receive such communication, are
entirely cut off from the life and thought and experience of the race embodied in
human words. Indeed, until they have been given civilised speech it is useless to
talk of continuing their education, for, in a real sense, their education has not yet
begun." (The Newbolt Report, quoted in Gordon 1981: 61)
It goes without saying that such allegations are nothing more than very thinly
disguised social prejudice, and they were promptly debunked (cf. Labov 1972, Gordon
1981). They had a profound influence on the history of linguistics, however, in that they
made it thoroughly unrespectable to espouse any form of the 'syntactic deficit theory'.22 It
was acknowledged, of course, that working-class dialects differed from those of the
middle classes, but 'different', it was pointed out, does not mean 'inferior'. From now on,
all the varieties of a given language were to be considered equal. At least in principle, for
if we look at the actual practice, we will soon discover that some turned out to be more
equal than others. Since the middle class dialect is the most accessible to middle class
researchers, it is this dialect which they study, and tacitly assume that whatever their
research reveals about the competence underlying their chosen variety will also be
applicable to other dialects. The problem with this approach is that the highly syntacticized
forms of the kind that they deal with simply do not occur in the speech of the working
classes (and are rare in the speech of the middle classes), and the types of texts which
31
require this style are not written in non-standard dialects. Therefore, syntacticians ended
up focusing primarily on the standard written language.23
Thus, the 'egalitarian' dogma, though highly commendable on moral grounds, was
hardly scientific. It lead to the systematic disregard of certain types of data, including data
about individual differences in the ability to make use of syntactic cues--which made the
nativist-modularist orthodoxy seem more plausible. It is also possible that it hurt those that
it was supposed to protect. Failure to automate the cognitive routines for processing
complex constructions is not likely to impair one's communicative potential in normal
everyday situations; it does, however, seriously impair one's ability to process highly
syntacticized written texts. To the extent that linguists have succeeded in persuading
teachers that syntax is basically innate and therefore pupils do not need any special
assistance in learning to make sense of the complex constructions that occur in some
written texts, they may have unwittingly closed the doors to higher education to those
who have not acquired that skill.
32
Appendix 1: The questionnaire
Q1.
Robert was convinced that Jerry would get the job even though he wasn't really
qualified for it. [control]
What was Robert convinced of?
What wasn't Jerry qualified for?
Q2.
Sandy will be easy to get the president to vote for.
Who will vote?
For whom will this person vote?
What will be easy?
Who will find it easy to do this?
Q3.
It was King Louis who the general convinced that this slave might speak to.
Who might the slave speak to?
Who did the convincing?
Who was convinced of something?
Q4.
The milkman who Paul saw after discovering the zoo-keeper planned to kill got off
the bus.
Who wanted to kill someone?
Whom did this person want to kill?
Who got off the bus?
Q5.
The manager knew that the fact that taking good care of herself was essential upset
Alice.
What did the manager know?
What upset Alice?
Q6.
That dog will be difficult to persuade the little girl to play with.
Who will play?
What will this person play with?
Who will find it difficult to do something?
What will be difficult to do?
33
Q7.
The nervous-looking student that Chris met after being told his girlfriend wanted to
jilt took the 11 o'clock train.
Who took the 11 o'clock train?
Who is likely to be jilted?
Who wanted to jilt someone?
Q8.
Alex decided that the easiest way to find out whether or not the plan would work
was to ask the man who played the guitar at the party. [control]
What did Alex want to ask about?
Why did she want to ask it?
Q9.
John remembered that the fact that criticizing himself was hard surprised Arthur.
What did John remember?
What surprised Arthur?
Q10. John will be hard to get his wife to vouch for.
Who will vouch?
For whom will this person vouch?
What will be hard?
Who will find it hard to do this?
Q11. It was the student counsellor that the dean convinced that the case should be turned
over to.
Whom did the dean manage to convince of something?
Who should the case be turned over to?
Q12. Paul noticed that the fact that the room was tidy surprised Shona.
What did Paul notice?
What surprised Shona?
Q13. Mary was well aware that, although everyone knew that the building had been
designed by John and herself, only he would receive the professional recognition that
would ensure his future in the field of architecture. [control]
Who designed the building?
Who was going to receive professional recognition?
34
Q14. The mayor who Julie warned after learning the ex-prisoner wanted to interrogate
managed to get away.
Who wanted to interrogate someone?
Who was supposed to have been interrogated?
Who managed to get away?
Q15. That little boy will be very difficult to persuade his teacher to criticise.
Who might criticise someone?
Who might be criticised?
What will be very difficult?
Who will find it difficult to do this?
Q16. It's the ticket conductor who Jim managed to convince that the red-nosed man was
looking for.
Who was the red-nosed man looking for?
Who convinced someone of something?
Who was convinced of something?
Q17. The story that we had to read after being told the general director wanted to give
first prize to was very boring.
Who was going to choose the winner of the prize?
Whom or what is the prize going to be awarded to?
What was very boring?
Q18. The criminal realized that the fact that leaving one's fingerprints on the revolver was
dangerous would not be overlooked by the investigators.
What did the criminal realize?
What was not overlooked by the investigators?
Q19. Jackie persuaded my mum to invite three famous novelists to participate in the
discussion programme, but they all declined. [control]
What did the three famous novelists decline to do?
Who invited someone to do something?
Q20. It was her boss that Louise persuaded that Margaret will interview.
Who will be interviewed?
Who persuaded someone of something?
Who was persuaded of something?
35
Appendix 2: Grammaticality/Acceptability judgements
Sentence type
Sentence number
Acceptability
rating
complex NP - 'difficult'
18
7.5
complex NP - 'neutral'
5
9.0
complex NP- 'neutral'
9
9.0
12
9.5
tough - 'difficult'
2
6.0
tough - 'neutral'
10
5.5
tough - 'neutral'
15
6.0
6
7.0
gap 1 - 'difficult'
20
4.0
gap 1 - 'neutral'
3
3.0
gap 1 - 'neutral'
16
4.0
gap 1 - 'easy'
11
5.5
gap 2 - 'difficult'
14
0.5
gap 2 - 'neutral'
4
2.5
gap 2 - 'neutral'
7
3.5
17
3.5
control
1
10.0
control
8
10.0
control
13
10.0
control
19
10.0
complex NP- 'easy'
tough - 'easy'
gap 2 - 'easy'
36
References
Brazil, David (1995). A Grammar of speech. Oxford: Oxford University Press.
Chien, Yu-Chin; and Wexler, Kenneth (1990). Children's knowledge of locality conditions
in binding as evidence for the modularity of syntax and pragmatics. Language
Acquisition 3, 225-295.
Chomsky, Noam (1980a). Rules and representations. Oxford: Basil Blackwell.
Chomsky, Noam (1980b). On Cognitive Structures and Their Development: A Reply to
Piaget. In Language and Learning. The Debate Between Jean Piaget and Noam
Chomsky, Piattelli-Palmarini, Massimo. (ed.), 35-52. London: Routledge and Kegan
Paul.
de Villiers, Jill G.; Tager-Flusberg, Helen; Hakuta, Kenji; and Cohen, Michael (1979).
Children's comprehension of relative clauses. Journal of Psycholinguistic Research 8,
499-518.
Ferreira, Fernanda; and Clifton, Charles, Jr. (1986). The independence of syntactic
processing. Journal of Memory and Child Language 25, 348-368.
Fletcher, Paul (1985). A Child's Learning of English. Oxford: Basil Blackwell.
Flynn, Suzanne; and Lust, Barbara (1980). Acquisition of relative clauses: Developmental
changes in their heads. Cornell Working Papers in Linguistics 1, 33-45.
37
Fodor, Jerry A. (1983). The Modularity of Mind. Cambridge, Massachusetts: MIT Press.
Forster, K. (1979). Levels of processing and the structure of the language processor. In
Sentence processing, Cooper, William E.; and Walker, Edward T. C. (eds.) Hillsdale,
New Jersey: Erlbaum.
Frazier, Lyn (1979). On comprehending sentences: Syntactic parsing strategies.
Bloomington: Indiana University Linguistics Club.
Geneshi, Celia; and Dyson, Anne Haas (1987 [1984]). Language assessment in the early
years. Norwood, New Jersey: Ablex Publishing Corporation.
Givón, Talmy (1979). From Discourse to Syntax: Grammar as a Processing Strategy. In
Discourse and Syntax. Syntax and Semantics 12, Talmy Givón (ed.), 81-112. New
York: Academic Press.
Goodluck, Helen (1990). Knowledge Integration in Processing and Acquisition: Comments
on Grimshaw and Rosen. In Language processing and language acquisition, Lyn
Frazier and Jill de Villiers (eds.), 369-382. Dordrecht: Kluwer Academic Publishers.
Goodluck, Helen (1991). Language Acquisition. A Linguistic Introduction. Oxford:
Blackwell.
Gordon, J. C. B. (1981). Verbal Deficit. A Critique. London: Croom Helm.
38
Grimshaw, Jane and Rosen, Sara Thomas (1990) Obeying the binding theory. In Language
processing and language acquisition, Lyn Frazier and Jill de Villiers (eds.), 357-367.
Dordrecht: Kluwer Academic Publishers.
Horgan, Diana (1978). The development of the full passive. Journal of Child Language 5,
65-80.
Hsu, Jennifer R.; Crains, Helen S.; Eisenberg, Sarita; and Schliesselberg, Gloria (1991).
When do children avoid backwards coreference? Journal of Child Language 18, 339353.
Hunt, Kellog W. (1977). Early Blooming and Late Blooming Syntactic Structures. In
Evaluating Writing, Ch. L. Cooper and L. Odell (eds.), 91-104. Buffalo, New York:
State University of Buffalo.
Labov, William (1972). The Logic of Nonstandard English. In Language in the Inner City.
Studies in the Black English Vernacular, William Labov, 201-240. Oxford: Basil
Blackwell.
Lakoff, George (1973). Fuzzy Grammar and the Performance/Competence Terminology
Game. Proceedings of the Ninth Annual Meeting, Chicago Linguistic Society, 271291.
Lightfoot, David (1991). How to Set Parameters: Arguments from Language Change.
Cambridge, Massachusetts: MIT Press.
39
Limber, John (1973). The genesis of complex sentences. In Cognitive development and the
acquisition of language, Moore, T. E. (ed.), 169-185. New York: Academic Press.
Maratsos, Michael; Kuczaj, Stan A.; Fox, Dana E. C; and Chalkley, Mary Anne (1979).
Some empirical studies in the acquisition of transformational relations: Passives,
negatives and the past tense. In Children's Language and Communication, W.
Andrew Collins, (ed.), 1-45. Hillsdale, New Jersey: Erlbaum.
Maratsos, Michael. (1984). The Acquisition of Syntax. In Handbook of Child Psychology,
vol. 3, 4th edition, John Flavell and Ellen M. Markman (eds.). New York: Wiley.
Marslen-Wilson, William; and Tyler, Lorraine Komisarjevsky (1987). Against Modularity.
In Modularity in knowledge representation and natural-language processing, Jay L.
Garfield (ed.), 37-62. Cambridge, Massachusetts: MIT Press.
Miller, J.E. (1994). Speech and Writing. In The Encyclopedia of Language and
Linguistics, vol. 8, R.E. Asher and J. M. Y. Simpson (eds.), 4301-4306. Oxford:
Pergamon Press.
Ochs, Elinor (1979). Planned and unplanned discourse. In Discourse and Syntax. Syntax
and Semantics 12, Talmy Givón (ed.), 51-80. New York: Academic Press.
Rayner, Keith; Carlson, Marcia; and Frazier, Lyn (1983). The interaction of syntax and
semantics during sentence processing: Eye movements in the analysis of semantically
biased sentences". Journal of Verbal Learning and Verbal Behavior 22, 358-374.
40
Shankweiler, Donald; Smith, Suzanne T.; and Mann, Virginia A. (1984). Repetition and
comprehension of spoken sentences by reading-disabled children. Brain and Language
23, 241-257.
Sheldon, Amy (1974). The role of parallel function in the acquisition of relative clauses in
English. Journal of Verbal Learning and Verbal Behavior 13, 272-281.
Tavakolian, Susan L. (1981). The conjoined-clause analysis of relative clauses. In
Language acquisition and linguistic theory, Susan L. Tavakolian (ed.), 67-87.
Cambridge, Massachusetts: MIT Press.
Tyler, Lorraine Komisarjevsky and Marslen-Wilson, William D. (1977). The on-line effects
of semantic context on syntactic processing". Journal of Verbal Learning and Verbal
Behaviour 16, 683-692.
41
Notes
1
Throughout this paper, I use the term 'generative' to refer to the research tradition which
includes Chomsky's original version of Transformational-Generative Grammar (TG) and
its various reincarnations, including the Starndard Theory, Governament-Binding (GB)
and a number of related frameworks.
2
These sentences are meant to be illustrations of the type of structures that one
encounters in generativist journals. My purpose is to provide examples of a type of
discourse rather than of the work of individuals: it is for this reason that no reference is
made to specific authors.
3
Due to the high unemployment which has persisted in the United Kingdom over the last
few years, some individuals have been forced to take on any jobs that were available,
rather than the jobs they were qualified to do. The least educated group did contain one
such person--a cleaner who had stayed at school till 18, had held a skilled job for almost
twenty years, and claimed to know Spanish. Not surprisingly, her scores on the
questionnaire were significantly higher than those of other members of the group.
4
Acceptability ratings for these structures obtained from the control group ranged from
0.5 to 4 on a scale of 0-10 (see footnote 8 for an explanation of how these figures were
arrived at).
5
To distinguish questionnaire items from linguistic examples, the former are marked with
the letter Q followed by the number of the test sentence as it appeared in the
questionnaire. (See Appendix 1 for a full list of sentences and questions.)
6
It seems that hearing the sentences helped all the respondents, including the most
literate ones: many of the lecturers explicitly stated this at the end of the interview.
42
7
Subjects were allowed to change their minds as many times as they wished; only the final
answer was scored.
8
These were used to calculate composite grammaticality judgements using the following
method: each time a sentence was classified as 'English', it got 1 point, and each time it
was classified as 'borderline', it got 0,5 point. Subsequently, the ratings from the ten
respondents were added up. This resulted in a grammaticality/acceptability scale ranging
from 0 (totally unacceptable) to 10 (perfectly normal). The composite ratings for each
sentence type are provided in Appendix 2.
9
Answers such as 'I don't know' and 'You can't tell' were scored as correct on this
question.
10
The problem is not actually as acute as the foregoing discussion suggests, since most
educated speakers gave 'correct' answers to most questions, and each test sentence was
judged at least partially grammatical by at least one informant.
11
The number of sentences without mistakes (rather than the number of sentences with
mistakes) were used as the basis of the B-score because this would yield a curve with a
similar shape to that obtained by applying method A.
12
Since one could obtain up to 4 points for each test question under scoring system A, the
B-scores have been multiplied by 4 to make the results comparable.
13
A more plausible 'performance' explanation for the differences between the results
obtained by the four groups in this study might be couched in terms of differences in
working memory capacity. To show that this is a relevant factor, one would have to
demonstrate (1) that there are in fact significant differences in working memory span
between the groups and (2) that the test sentences require greater working memory
resources than the control sentences. Of course, this would still not constitute proof that
the poorly-performing groups do in fact know the test structures.
43
14
It is no good arguing that triggering data for the acquisition of these structures occurs
only (or primarily) in formal discourse, and hence some respondents may not have been
exposed to the relevant data. All the subjects of this study have been exposed to formal
English for a number of years; if this is not enough then the nativist position becomes
indistinguishable from the empiricist position. One could, of course, claim that the
particular structures used in this study are extremely rare, and hence many respondents
may not have received adequate exposure -- but this suggests that perhaps generative
linguists should base their sweeping claims on more mundane data. However, it will
certainly be worth repeating the study using more natural material.
15
It is also worth noting that subjects tended to respond with the first NP in the sentence
even when this resulted in a highly implausible response. For example, when presented
with Q6, That dog will be difficult to persuade the little girl to play with, and asked what
will be difficult, many respondents said, correctly, 'persuading the little girl to play with the
dog'. When they were later asked who will find it difficult to do this, the overwhelming
majority of the least educated subjects said 'the dog'.
16
Responses such as 'that criticising himself was hard and that Arthur was surprised' were
counted as correct and have been included in the 'that Arthur was surprised' category.
17
One might object to this claim by pointing out that unplanned informal discourse poses
processing difficulties of a different kind--for example, reconstruction of ellipted
constituents, finding antecedents for anaphora, matching filler and gap in questions and
topicalized constructions. However, we do not know to what extent these processing
problems are resolved by using syntactic strategies; it may well be that listeners rely
primarily on contextual information and real-world knowledge. Be that as it may,
constructions such as those studied here simply do not occur in ordinary conversation.
18
For a discussion of the role of entrenchment in language processing, see Langacker
1987, 1991. Briefly, the concept helps elucidate the different senses in which one may
44
'know' the procedure for assembling (or analyzing) a composite linguistic structure. For
one speaker, the procedure may be easily accessible as a prefabricated unit (and hence
require minimal effort to process), while another speaker may command only the
elementary building blocks and need to build up the composite structure from scratch; in
this case, processing might require considerable effort. Although both speakers may be
said to 'know' the structure, calling up a well-practiced routine is very different from
deriving it from first principles.
19
It is also possible the language user lacks the syntactic routines necessary for processing
a certain type of sentence, in which case he or she will have to rely on other sources of
information in order to make any sense of it.
20
It is worth pointing out that Fodor's argument is based on the assumption that 'higher
level' processes such as inferencing are always slower than modularized processes.
Research by Marslen-Wilson and Tyler (Tyler & Marslen-Wilson 1977, Marslen-Wilson &
Tyler 1987), however, shows that this is not the case, for very common syntactic
structures (for which we presumably develop highly automated routines).
21
Of course it does not follow from this that human linguistic abilities have no innate
component whatsoever. What I am claiming here is that some abilities which are assumed
to to fall out naturally from the biological endowment are in fact learned--learned in the
step-by-step empiricist way, not 'acquired'.
22
Bernstein's repeated denials that his theory is a verbal deficit theory is a good illustration
of this.
23
The above argument, of course, is an oversimplification. It is not entirely true that the
highly syntacticized forms that we are interested in appear only in written texts. Although
most typical of written discourse, they spill over into spoken language, but it is the spoken
language of those who read a great deal--i.e., the middle classes. They are not very likely
to occur in the speech of unskilled labourers.
© Copyright 2026 Paperzz