Applied Linguistics Advance Access published January 17, 2011
Applied Linguistics: 1–25
doi:10.1093/applin/amq053
! Oxford University Press 2011
A Multidimensional Analysis of a
Written L2 Spanish Corpus
*YULY ASENCIÓN-DELANEY and JOSEPH COLLENTINE
Northern Arizona University, Modern Languages, Box 6004, Flagstaff, AZ 86011, USA
*E-mail: [email protected]
INTRODUCTION
Second language acquisition (SLA) research views learning not only as knowing individual lexical and grammatical features of a second language (L2) but
also how these features work together to generate types of discourse (e.g.
narratives, descriptions). In discussing certain mismatches between linguistic
theories, psycholinguistics, and corpus-based investigations, Ellis (2002) comments that there has been traditionally a ‘blind faith’ in the (psycholinguistic)
reality of structural categories (e.g. noun, verb). Collentine and AsenciónDelaney (2010) use corpus-based techniques to show that, statistically speaking, seemingly disparate lexical and grammatical L2 structures (e.g. verbs and
adjectives) develop in tandem and work together to affect different types of
discourse. Indeed, as Ellis et al. (2008) note, there is evidence that ‘human
production grammar must store probabilistic relations between words’
(p. 377). Furthermore, Ellis et al. (2008) postulate that an important difference
between native and non-native speakers is that ‘native speakers have extracted
underlying co-occurrence information’.
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
The present study adds to our understanding of how learners employ lexical and
grammatical phenomena to communicate in writing in different types of interlanguage discourse. A multidimensional (factor) analysis of a corpus of L2
Spanish writing (202,241 words) generated by second- and third-year,
university-level learners was performed. The analysis uncovered four significant
clusters that can be considered distinct discourse types with two main stylistic
variations: narrative (characterized by verbal features) and expository (characterized by nominal features). Results also provide examples of the multiple ways
that stylistic sophistication and linguistic complexity occur in the L2. Although
the Spanish learners’ discourse did not show signs of syntactic complexity (e.g.
frequent use of relative clauses, subordinate clauses, use of clitics), the frequent
use of nominal features affects informational density due to the presence of
numerous derivational morphemes. Inflectional complexity in the form of
marked forms was not predominant in the data set. Still, the learners’ verbal
inflections did vary, which is a sign of L2 development (Howard 2002, 2006;
Collentine 2004; Marsden and David 2008).
2 ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
Developing L2 discursive abilities
Canale and Swain’s (1980) and Bachman’s (1990) theories of L2 communicative competence afford discourse/textual competence a central role in SLA.
The American Council on the Teaching of Foreign Languages (ACTFL 2001)
proficiency scales assume a developmental progression, with learners developing conversational, then descriptive, then narratives abilities and finally more
complex abilities. In general, the study of discursive abilities entails how learners obtain the ability to produce multipropositional segments whose semantic
and structural interrelationships (e.g. gender/number agreement) cohere
(Givón 1985). This research has considered how pragmatic abilities affect L2
discursive abilities, the assignment of discursive roles to grammatical constructs, and learners’ use of discourse markers.
Pragmatic research concerned with the discursive context studies how we
structure information across sentences, treating issues such as how syntactic
structures affect coherence, topic/focus, and propositional presupposition
(Horn and Ward 2006: xiii–xix). SLA pragmatic research, as Kasper (2001)
notes, examines how learners organize discourse and how they employ grammatical structures to produce multipropositional messages (e.g. marking
cause-effect chains with conditional if structures). The acquisition of what
may be termed ‘grammatical-pragmatic’ abilities entails not only lexical and
grammatical phenomena’s denotations but also their connotations as managed
by readers/listeners to interpret multisentential messages.
Research suggests that L2 discursive abilities develop as learners attain both
grammatical and pragmatic abilities. In uninstructed learning contexts, functional necessities (e.g. distinguishing between topic and comment) influence
which lexical and grammatical tools learners grammaticalize (Skiba and
Dittmar 1992). Geyer (2007) presents data suggesting that once Japanese
learners develop ‘several (grammatical-pragmatic) devices’ (p. 362) they can
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Researchers have powerful analytical tools for studying how lexical and
grammatical features generate discourse types (Biber and Conrad 2001).
Still, Myles (2005) and Myles and Mitchell (2005) argue that SLA research
has been slow to embrace corpus-based technologies even though corpus linguistics is effective at helping theoretical linguistics to understand the nature of
discourse types. Myles and Mitchell (2005) also note that corpus research can
complement and increase the generalizability of SLA research by examining
large amounts of data.
We present the first-known multidimensional analysis of a written Spanish
learner corpus examining how learners combine lexical and grammatical phenomena to generate different discourse types. Before presenting our research
questions, we provide an overview of research about the acquisition of discursive abilities, how corpora have been used in SLA to understand the mapping
of form to function, and the usefulness of the multidimensional corpus
analysis.
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
3
SLA corpus research on the development of structural and
lexical phenomena
Corpus-based L2 research has helped us to understand how learners map form
to function.2 It also elucidates the role of L2 formulaic segments. It has provided important insights into the interaction of grammar and the lexicon.
What is consistent within this research is that L2 lexical phenomena and grammatical constructs work in tandem. Finally, corpus-based L2 research has
explored ways to efficiently measure L2 syntactic complexity.
Regarding the mapping of form to function, Klein and Perdue (1997) used a
learner corpus to posit that uninstructed settings generate a ‘basic variety’,
where word order and grammatical constructs result from functional communicative necessities. Granger et al. (2002) and Belz (2004) use corpus techniques to document how social and institutional pressures affect new L2
phenomena (e.g. da-compounds such as dazu ‘there-to’, davon ‘there-from’,
darin ‘there-in’).
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
chain utterances to achieve discursive organization. Geyer reports that
advanced learners use coordinate and adverbial conjunctions to make
distinctions, such as foreground/background and cohesive propositional
segmenting.
The research on the acquisition of Spanish’s two copulas ser and estar shows
how learners associate lexical and grammatical features with particular discourse types. Cheng et al. (2008), examining a corpus of Mandarin Chinese
L1 learners of Spanish, report that exploratory writing led to more copula
estar + adjective usage. They also noted that this structure was mostly associated with narratives and descriptions. Collentine and Asención-Delaney
(2010), in their corpus-based study of the Spanish copula ser + adjective and
estar + adjective, report that overall, when learners begin mastering simple discourse types, lexical, syntactic, and morphological diversity increases and complexity increases.1 They found that ser + adjective appears in descriptive and
evaluative discourse where lexical and morphological diversity and complexity
occurs. However, estar + adjective is present in narrations, descriptions, and
hypothetical discourse where, nonetheless, surprisingly little linguistic complexity occurs.
Other discourse related research has studied L2 discourse markers. Upton
and Connor (2001) use corpus techniques to compare formulaics in ‘application letters’ by L1 speakers of (American) English and L2 learners of English
from Belgian and Finland. The L2 writers were much less formulaic than the
L1 writers with politeness strategies, due to L1 influence. Forsberg (2005) uses
an L2 French learner corpus from L1 Swedish speakers to study different types
of idioms (e.g. grammatical, pragmatic, purely lexical idioms). Forsberg found
that lexical idioms increase in writing over time. She also found that formulaics
with a discursive role were ‘overused’ by learners of all levels.
4 ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Some investigators conceptualize formulaic segments (e.g. in other words, for
example) as singular dictionary entries (cf. Ellis 2006). Using the British
National Corpus (BNC Consortium 2001), Ellis et al. (2008) found that L2
learners at early stages of acquisition recognize formulaics based on their frequency in the input. Interestingly, however, Ellis et al. (2008) argue that
native-like use of formulaics ultimately requires learners to alter their processing of words ‘for coherence: for co-occurrence greater than chance’ (p. 391).
Regarding the interaction between morphology and the lexicon, Marsden
and David (2008) study vocabulary development among L2 learners of Spanish
at ages 9 and 13. The 13-year-old learners, who had 450 more hours of classroom instruction, showed greater lexical and inflectional diversity. Their analysis supports the notion that, in inflectionally rich languages like Spanish, an
important indicator of development is not only accuracy, as measured by
grammatical errors, but also increased inflectional/derivational variation
(Howard 2002, 2006; Collentine 2009). Their research also indicates that a
key factor in differentiating levels of development is changes in the part of
speech (e.g. nouns, verbs, adjectives) that predominate in production.
Marsden and David (2008) present evidence suggesting that as learners progress, in speech they produce more verbs than nouns, and that as they begin to
produce more verbs, they also start to produce more adjectives. Since learners
produce different discourse types as they progress, it is logical to surmise that
these changes in lexical and grammatical production parallel changes in the
discourse learners’ produce.
Two research projects have used corpus techniques to study the emergence
of linguistic complexity and semantic density. Operationalizing ‘complexity’
and ‘density’ in corpus-based studies is challenging but important. Ortega
(2000) uses a corpus of intermediate-level L2-Spanish learners’ writing to
identify reliable measures of syntactic complexity. Her analysis indicated that
clause length, phrasal elaboration, and amount of subordination best predict
syntactic complexity. Collentine (2004) employs corpus techniques to compare
morphological and lexical complexity in an in-class learning context and a
study-abroad context. After a semester, the study-abroad learners employed
more morphological narrative complexity by using past-tense verbs, thirdperson morphology, past participles, and present participles (as well as public
verbs; e.g. decir que ‘to say that’). From a learner’s perspective, morphological
complexity in Spanish requires the use of a range of aspectual, person,
number, and gender inflections beyond simple verb tenses (e.g. the present)
and other unmarked morphemes (e.g. masculine-singular nouns/adjectives; cf.
Collentine 2004: 237). The in-class group was more lexically complex, producing a higher concentration of nominal features—nouns and adjectives—and
so semantically dense discourse (Biber 1988).
Finally, Grant and Ginther (2000) urge corpus-based SLA researchers to
combine qualitative with quantitative techniques to increase generalizability
of results. Many approaches (e.g. part-of-speech tagging) do not generally
factor in errors and small corpora bias from sampling particular tasks. While
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
5
we do not report error rates or omissions in this study, we are confident that
our multidimensional approach and its mixed method research design (i.e.
quantitative + qualitative analyses), together with the size of our corpus, address Grant and Ginther’s (2000) concerns.
The multidimensional analysis and relevant corpus linguistic
techniques
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Multidimensional corpus analysis shows how lexical and grammatical features
bundle together to produce different and new types of discourse (Biber and
Conrad 2001). It combines technological tools, exploratory factor analysis, and
qualitative analysis of texts to determine which lexical and grammatical features reliably co-occur in a large corpus. Using a large collection of texts
coupled with powerful statistical tools introduces fewer reliability threats
than small-scale studies, where participants’ particular characteristics or
raters’ judgement heavily influence results.
Biber et al. (2006) provide the first multidimensional analysis of nativespeaker Spanish. They analyze a 20-million-word Spanish corpus (4,049
texts) with written and oral data representing 19 registers (e.g. face-to-face
conversation, business letters). The analysis uncovered both well-known discourse types like narratives as well as new ones. For example, the following
characterize ‘informationally rich’ discourse in Spanish: nouns, adjectives, definite articles, prepositions, derived nouns, type-token ratio, long words
(i.e. multisyllabic words), and ergative se constructions. Unique to Spanish is
also hypothetical discourse, containing concentration of structures such as the
conditional and the subjunctive as well as future verb forms, verbs of obligation and causation (e.g. dejar, permitir, hacer + infinitive) and the conjunction
que.
Parodi (2007) used a 2.5-million-word corpus of native-speaker Spanish to
study the differences between written and spoken Spanish. His analysis complements Biber et al. (2006) in that it reveals how lexical and grammatical
features cluster together, based on whether the register is context dependent
(i.e. the interpretation of important features depends on the ‘speech situation’), written academic, commissive in nature (in the pragmatic sense), attitudinal, or informational in focus. Additionally, Parodi provides a more
refined definition (although based on many fewer tokens) of what constitutes
Spanish narrative discourse than Biber et al. (2006); interestingly, Parodi’s
analysis indicates that English and Spanish share many of the same narrative
features.
We present the first-known multidimensional analysis of L2 Spanish, studying how learners use various lexical and grammatical phenomena to generate
IL-specific discourse types. We do not make a priori assumptions about which
of these phenomena work in tandem nor do we assume that learner discourse
types are the same as those of native-speakers. Our analysis serves as a first
6 ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
step towards characterizing learner discourse at the second and third years of
Spanish L2 instruction.
Research questions
We characterize Spanish learner discourse via a multidimensional analysis
of a corpus of L2 Spanish generated by second- and third-year, universitylevel learners. Specifically, the study addresses the following research
questions:
METHOD
Corpus description
We used a 202,241-word corpus of written Spanish, comprising edited and
non-edited compositions collected from English-speaking Spanish learners at
the second-year (109,224 words) and third-year (93,017 words), levels in
which more variety of texts could be collected due to more exposure to language instruction. To estimate the written proficiency of each level of instruction based on the ACTFL Writing Proficiency Scale (ACTFL 2001), the
researchers selected a random sample of 50 entire documents from each of
the two instructional levels (N = 100). After a training session on working with
the scale, the researchers rated the samples independently. Each level on the
scale was assigned a numerical value, with 0 representing a novice low and 9 a
superior rating. We estimated the inter-rater reliability of the subjects’ proficiency ratings with a Pearson correlation since the datasets represented interval scales [r(df = 98) 0.97; p < 0.01]. The second-year learners wrote at the
intermediate high level and the third-year learners at the advanced low
level. This suggests that while the third-year learners generally narrated and
produced a limited number of cohesive devices as well as a variety of complex
syntactic structures, the second-year learners were beginning to produce narrative structure, although their control of the verbal constructs was still developing. The second-year learners also produced few cohesive devices and
limited subordination.
The corpus comprises writing samples used for course assessment purposes:
letters, narratives, descriptions, summaries, and argumentative essays, both in
and out of class as well as on exams. Given the study’s exploratory and descriptive nature, we did not control the type of tasks or topics within the
corpus. Topics related to textbook themes (e.g. family, childhood) and cultural
readings.
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
1 How do lexical and grammatical phenomena cluster together in L2
Spanish writing?
2 What types of discourses (e.g. narrative, descriptive, hypothetical) surround the concentration of these lexical and grammatical features?
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
7
Procedures: tagging, searching and norming
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Understanding multidimensional analyses requires knowledge of three key
concepts: part-of-speech tagging, search pattern, and statistical techniques
for eliminating various biases in how tokens are counted.
To access information about particular texts, each file includes a header with
information about its topic, source type, author biographical information, and
purpose (argumentative essay, narrative). To search for morphosyntactic information (e.g. all adjectives, all subjunctive forms), one needs a part-ofspeech tagger: software that annotates every word with information about
its major word classes (e.g. adjective, noun, verb, determiner), basic morphological information (e.g. plural, preterit), as well as its lemma (i.e. its unmarked, dictionary root, such as a verb’s infinitive or a noun’s masculine,
singular form).
Part-of-speech tagging requires a dictionary with lexical and grammatical
information. It also requires a pretagged corpus to train the software routines
to determine unknown or ambiguous words’ probable tags. We compiled our
own dictionary, utilized a training set from samples from the Corpus del español
(Biber et al. 2006), and tagged the corpus with n-gram software routines from
the Natural Language Tool Kit (NLTK; http://www.nltk.org/). Additionally, we
wrote Spanish-specific routines (e.g. clitic sequences, derivational morphemes)
to complement the NLTK routines to achieve greater tagging precision. After
the corpus is tagged in this way, the investigator must verify the accuracy of
the tagging and fix errors through further programming.
We studied 78 lexical, grammatical, and lexico-grammatical features.3 The
features involved all parts of speech, common morphosyntactic constructs
studied by learners, as well as additional constructs studied in Biber et al.
(2006). They represented adjectives (e.g. derived, postnominal position),
nouns (e.g. derived, feminine), adverbs (e.g. place, time), verb classes (e.g.
imperfect aspect, past participle), verb phrases (e.g. communication, knowledge), and certain morphosyntactic features like dependent clauses, noun
phrase configurations (e.g. article plus noun), and pronoun usage (e.g.
clitic—third person).
Textual frequencies often require two mathematical conversions. First,
norming transforms a phenomenon’s count to its normed frequency. Since text
lengths vary, longer texts inflate certain items’ importance. To offset this
text-length bias one scales a phenomenon’s frequency per text, such as per
1,000 words. Second, normalizing eliminates the feature-concentration bias:
some phenomena are naturally scarce in a document (e.g. the subjunctive)
while others are naturally common (e.g. articles) (cf. Biber and Conrad 2001).
Normalizing converts a normed frequency to its z-score value vis-à-vis its
normed frequency in each document. Consequently, one can measure the
relative presence of two or more linguistic features within any given text.
Additionally, one can sum various z-scores to determine how concentrated a
set of features occur in any text or group of texts.
8 ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
Factor analysis
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
To determine how learners cluster lexical and grammatical features, we
employ an exploratory, principal factor analysis (PFA) (Biber and Conrad
2001). PFA identifies dimensions—also referred to as factors—along which
the 78 features in Table 1 co-vary statistically. The result is a series of dimensions according to which one could classify the texts of a corpus and the features characterizing each dimension. Frequently, a factor may have two
opposing clusters, which is why this technique is often referred to as a multidimensional analysis. Opposing clusters of features occur in complementary
distribution within the corpus.
The factors differ in terms of how much variation they account for. The
dimension accounting for most variance represents the primary, superset
factor according to which all texts can be classified. Dimensions that represent
less variance are subset factors of the superset. A superset factor may identify
the features of formal and informal language, with the remaining factors representing genres within the superset. Each factor has an eigenvalue, which
represents its importance relevant to the others. The examination of factors’
eigenvalues in scree plots of each factor’s total variance helps to determine
how many factors are important enough to report. In a scree plot, the eigenvalues for each factor typically flatten out at some point, which is an indication
of relatively unimportant factors. The measurement of a feature’s importance
within a factor is referred to as its loading, which takes the form of a correlation
coefficient that represents how much the feature correlates with the cluster
identified, varying from 0.00 or no correlation to !1.00 or absolute correlation.
A dimension with two significant clusters will have two sets of loadings, a set of
positive loadings and another set of negative loadings, differing mathematically
in terms of their sign (although the direction of a cluster’s sign is irrelevant).
Higher absolute loadings are more useful in the interpretation of a cluster’s
communicative function. For a cluster to represent some meaningful discourse
type, Biber (1988) recommends that it should have at least five loadings at or
above the !0.30 cutoff. What is useful about this approach is that it identifies
different discourse types in a corpus, and the loadings reveal the features that
most represent those discourse types. Biber et al. (2006), for instance, found six
dimensions in oral and written native-speaker Spanish: four had significant
clusters of both positive and negative features; two had only one significant
cluster. In the first (superset) dimension, the positive features (e.g. nouns,
post-modifying adjectives, long words, high-type token ratio) represented semantically dense, literate discourse and the negative features’ (e.g. suasive
nominal clauses, second-person pronouns, demonstratives) oral discourse.
To obtain a factor analysis that includes the fewest, most representative
features in each factor while considering idiosyncrasies of textual data, linguistic PFA results are commonly ‘rotated’ (Biber 1988). This mathematical technique maximizes high loading values and minimizes low values. Following
Biber (1988), we rotated the factors using the Promax method.
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
9
Table 1: Targeted linguistic features
Variable
Noun
Derived nouns (NOUN_DER)
Feminine nouns (NOUN_FEM)
Masculine nouns (NOUN_MSC)
Non-pluralizing nouns, e.g. virus ‘virus’
(NOUN_NPL)
Plural nouns (NOUN_PLR)
Singular nouns (NOUN_SNG)
Apocopated adjectives, e.g. buen ‘good’
(ADJ_SHRT)
Derived adjectives (ADJ_DERV)
Feminine adjectives (ADJ_FEM)
Four-inflection adjectives (ADJ_TYP1)
Masculine adjectives (ADJ_MSC)
Plural adjectives (ADJ_PLR)
Postmodifying adjectives (ADJ_POST)
Premodifying adjectives (ADJ_PRE)
Singular adjectives (ADJ_SNG)
Two-inflection adjectives (ADJ_TYP2)
Third-person clitics (PRON_3RD)
Preverb clitics (CLT_PRE) se + 3s verb, e.g. se
come ‘one eats’ (SE_VRB3S)
Subject pronouns (PRON_SUB)
Unplanned se (SE_UNPLANNED)
Article—noun segments (ART_NOUN)
Definite articles (DEFART)
Possessive adjectives (DET_POSS)
Possessive syntax (WO_POSS)
Present participle modified by noun, e.g. agua
hirviendo ‘boiling water’ (PS_NOUN)
Third-person verb (VRB_3RD)
Conditional (VRB_COND)
Copula plus adjective (not in other predicate
categories) (COP_ADJ)
Evaluative predicates (PRD_EVAL)
Future (VRB_FUT)
Gustar-like verbs (VRB_GULI)
Imperfect (VRB_IMP)
Infinitive (VRB_INF)
Adjective
Pronoun
Other noun phrase elements
Verbs
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Variable class
10
ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
Variable class
Miscellaneous
Infinitive not preceded by verb or article
(VRB_INVA)
Past participle (VRB_PS)
Past perfect (VRB_PPRF)
Past subjunctive (VRB_PSBJ)
Perfect tense verb (VRB_PRFC)
Periphrastic future, e.g. voy a comer ‘I’m going
to eat’(VRB_PERI)
Predicates of conclusions. e.g. es obvio ‘it is
obvious’ (PRD_CONC)
Predicates of probability (PRD_PROB)
Present participle (VRB_PP)
Preterit (VRB_PRET)
Progressive aspect (VRB_PRGA)
Saber/conocer (SAB_CON)
Suasive predicates (PRD_SUAS)
Suasive verbs (VRB_SUAS)
Subjunctive (VRB_SBJ)
Verbs of communication, reporting
(VRB_COM)
Verbs of conclusions (VRB_CONC)
Verbs of knowledge; cognitive verbs
(VRB_CONO)
Verbs of observation (VRB_OBS)
Verbs of probability (VRB_PROB)
Adverb of time (ADV_TIME)
Adverbs of intensity (ADV_INTS)
Adverbs of manner (ADV_MAN)
Adverbs of place (ADV_PLC)
Adverbs of probability (ADV_PROB)
Comparatives (COMPARE)
Conjunctions followed by finite verb
(CONJ_VRB)
Irregular comparatives (COMP_IRR)
Lexical density ratio (DEN_TOK)4
Long words (LONG)5
Prepositions (PREPS)
Que subordinator (QUE)
Type token ratio (TYP_TOK)
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Adverbs
Variable
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
11
Variable
Adverbial clauses
Adverbial conjunctions of mode, e.g. lo hice
según me dijeron ‘I did it the way they told
me’ (ADVCLS_M)
Adverbial conjunctions of place, e.g. la casa
donde vivo ‘the house where I live’
(ADVCLS_L)
Adverbials clauses of time (ADVCLS_T)
Causal adverbial conjunctions (ADVCLS_C)
If clauses (ADVCLS_S)
Purpose adverbial clauses, i.e. para que ‘so
that’ (ADVCLS_P)
Relative pronouns with pre-posed preposition
(REL_EN)
Relative pronouns with pre-posed preposition
(REL_NO_EN)
Relative clauses on subjects, e.g. la casa que
está allá ‘the house that is there’(REL_SUBJ)
Nominal clauses—non-subjunctive
(NOM_IND)
Nominal clauses—subjunctive (NOM_SUBJ)
Adjective clauses
Nominal clauses
A factor’s meaning requires interpretation: What linguistic activity does this
cluster of features represent? To aid in this interpretive process one sums the
z-scores of a cluster’s significant loadings in order to easily identify the most
representative texts of a factor or cluster. Then, in the qualitative analysis of
our study, we examined the most representative texts using Biber et al. (2006)
and other researchers’ (e.g. Longacre 1983) description of Spanish discourse
types in order to determine the type of discourse portrayed by the cluster of
linguistic features in those texts (see Figure 1 for distribution of words in the
corpus by discourse type).
RESULTS
The following section summarizes the findings for the multidimensional analysis. First, we identify the five dimensions yielded by the statistical analysis.
For each dimension, we present the lexical and grammatical features clustering
for positive and negative loading sets of the dimension, and their communicative functions. Finally, differences between second and third year in each
dimension are addressed.
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Variable class
12
ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
Factors and dimensions
The scree-plot analysis of the eigenvalues for the rotated factors of the PFA
indicated that a five-factor solution was optimal, representing 20.2% of the
shared variance (Table 2).
Table 3 shows the feature clusters in the five factors with their loadings
meeting the !0.30 cutoff.
The following details the findings and our interpretations of the clusters’
discursive function, along with qualitative analysis of representative samples
of the factors.
Table 2: Eigenvalues of the rotated factor analysis
Factors
Eigenvalues
Percent of variance
Cumulative (%)
1
2
3
4
5
4.459
3.691
2.697
2.561
1.789
5.945
4.922
3.596
3.415
2.386
5.945
10.867
14.463
17.877
20.263
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Figure 1: Corpus word counts by discourse type
VRB_IMP (0.56)
VRB_PRGA (0.55)
VRB_PRET (0.53)
VRB_SBJ (0.42)
PRON_3RD (0.35)
VRB_CONO (0.35)
VRB_PSBJ (0.34)
VRB_COND (0.33)
CLT_PRE (0.33)
SAB_CON (0.30)
Negative cluster (loadings) ART_NOUN ("0.65)
NOUN_FEM ("0.56)
NOUN_DER ("0.53)
NOUN_PLR ("0.51)
NOUN_MSC ("0.36)
ADJ_TYP1 ("0.36)
ADJ_POST ("0.35)
DEFART ("0.31)
NOUN_SNG ("0.31)
ADJ_FEM ("0.30)
1
ADJ_DERV (0.71)
ADJ_PLR (0.59)
ADJ_SNG (0.52)
ADJ_TYP1 (0.51)
ADJ_MSC (0.47)
TYP_TOK (0.44)
ADJ_TYP2 (0.43)
ADJ_POST (0.41)
ADJ_FEM (0.35)
LONG (0.30)
NOUN_SNG ("0.40)
DEN_TOK ("0.38)
2
5
VRB_INVA (0.36)
VRB_INF (0.35)
4
PREPS (0.53)
VRB_INVA (0.52)
VRB_INF (0.49)
ADJ_PLR (0.31)
VRB_IMP ("0.45)
ADJ_SNG ("0.59) DEFART ("0.53)
VRB_PRGA ("0.42) ADJ_MSC ("0.44) ADJ_PLR ("0.40)
PREPS ("0.30)
VRB_3RD (0.63)
QUE (0.55)
SAB_CON (0.43)
VRB_CONO (0.42)
REL_NO_EN (0.38)
VRB_PROB (0.30)
3
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Positive cluster (loadings)
Factor
Table 3: Summary of factors (i.e. dimensions) and significant clusters
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
13
14
ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
Factor 1: Narrative vs. expository prose
Narrative. Era un dı́a lleno de sol y el aire lleno de aromas diferentes a flores
silvestres.. . . Al salir de mi dormitorio, todo parecı́a normal. No habı́a ninguna
nube que se pudiera ver en el cielo, entonces después de hacer esta observación continué con mi plan de salir de mi dormitorio. Comencé a correr por los caminitos
designados, y como el dı́a todavı́a parecı́a muy bonito decidı́ meterme adentro del
bosque. (It was a sunny day and the air filled with aromas of different wildflowers.. . . Upon leaving my room, everything seemed normal. There was no
cloud that could be seen in the sky, then after making this observation I
continued with my plan to leave my bedroom. I started to run around the
designated trails and as the day still looked very nice I decided to get into the
woods.)
Figure 2 describes the text types that have the highest average summed
z-scores for the features in the narrative cluster. Text types with the highest
average summed z-score are most representative of the cluster; while those
texts with the highest opposing (i.e. the opposing sign) are antithetical to the
cluster’s function.
Interestingly, not only do narrative texts have high positive scores in dimension 1, but also argumentative essays and summaries. Students frequently
approached summaries and argumentation with narrative elements; perhaps
to compensate for a lack of more sophisticated abilities, which has shown to be
the case for L2 writers of English in expository writing (Hinkel 2004). For
instance, one second-year learner wrote an argumentative text about stereotypes of the colonial times and used evidence from a story about a fray in the
new land.
Argumentative essay. El cuento de fraile Bartolomé y los indı́genas maya es una
estereotipo de los tiempos de la conquista de América Central. También las relaciones
malas entre las culturas europeas y indı́genas. El cura fue a guatemala para convertir a
los indı́genas. El misionero Bartolomé pensaba que su cultura de españa serı́a más
avanzado que los maya. En última instancia este pensamiento fue su último. (The
story of Brother Bartholomew and the indigenous Maya is a stereotype of the
time of the conquest of Central America. Also, the bad relations between indigenous and European cultures. The priest went to Guatemala to convert the
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Factor 1 contained two significant clusters. The analysis indicates that secondand third-year learners’ written production tends to be either narrative or
expository.
The lexical and grammatical features in the positive set include eight verbal
and two pronominal features, which constitute a narrative discourse. Two of
the verbal features with the largest loadings (i.e. imperfect and preterit) are
grammatical features used to present events and background descriptions in
the past (Biber et al. 2006). Also, third-person pronouns and clitics are used to
refer to presupposed participants or story protagonists in the narratives
(Longacre 1983).
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
15
Indians. The missionary Bartolome thought that his culture from Spain
would be more advanced than the Mayan. Ultimately this was his last
thought.)
In these narratives, the writers are present and involved. The cluster has
grammatical variables including verbal features such as subjunctive, past subjunctive, conditional, and progressive aspect, the first two of which represent
‘irrealis’ modality (Biber et al. 2006), where learners describe feelings and
attitudes towards possible eventualities. The presence of a lexical feature,
such as knowledge verbs, indicates the writer’s perceptions (Weber and
Bentivoglio 1991). These features personalize the narratives. Finally, it is
also noteworthy that the presence of subjunctive forms in narratives indicates some degree of syntactic complexity in the form of subordination
(Parodi 2007).
The third-year learners trended towards using more narrative behaviors,
whereas the third-year learners averaged a higher summed z-score on this
cluster (M = 4.1; SD = 5.9) than the second-year learners (M = 3.0; SD = 5.9);
the difference is notable but did not approach significance [F(1, 634) = 3.7;
p = 0.06].
Dimension 1’s negative cluster included, exclusively, features associated
with noun phrases. Discourse concentrated with nominal features such as
nouns, articles, and adjectives is largely expository, where information is
condensed into a few words (Biber et al. 2006). Such semantically dense discourse is often achieved through the use of multiple derivational morphemes
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Figure 2: Concentration of narrative features by text type and averaged
summed z-score
16
ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
(Biber 1988; Collentine 2004). Interestingly, learners in their second and third
years already possess their own strategies for producing ‘literate’ discourse,
some of which overlap with native-speaker features (cf. Biber et al. 2006).
Figure 3 reports the average summed z-score of the expository features by
text type, confirming our interpretation that the negative cluster of dimension
1 represents expository discourse.
The expository cluster also occurs in mini-essay questions and descriptions,
which the Expository essay section exemplifies.
Expository essay. Los Estados Unidos es un paı́s bastante rico y los
jóvenes tienen bastantes oportunidades en la vida. No obstante, la
mitad de los jóvenes del Estados Unidos son malsanos y más y más
tienen los problemas con hiperactividad y aprendizaje. Es obvio que
muchos jóvenes están faltando los elementos necesarios del contentamiento y una vida saludable. (The United States is a rather rich country
and young people have many opportunities in life. However, half of the youth
of the United States are unhealthy and more and more have problems with
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Figure 3: Concentration of expository features by text type and averaged
summed z-scores
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
17
hyperactivity and learning. It is clear that many young people are lacking the
necessary elements of a happy and healthy life.)
The third-year learners produced more nominal discourse features. The
third-year learners averaged a higher summed z-score on this cluster
(M = 2.4; SD = 4.7) than the second-year learners (M = 0.5; SD = 5.2), with
the difference being statistically significant [F(1, 634) = 16.3; p # 0.00].
Factor 2: Descriptive expository prose
Descriptive expository essay. El sueño americano se infiltra desde juventud, es evidente por televisión, escuela, la cultura y ejemplos del gobierno
y polı́ticos. Esta influencia es subconsciente pero fuerte y se enseña el americano que la única cosa que se necesita hacer es trabaja fielmente y comprar
las cosas correctas y eventualmente se recibirá la vida perfecta. Entonces,
en la mente americano la definición del suceso es ser rico, ser bonito y
joven y tener un montón de las cosas mejores y ricas. [The American
dream infiltrates from youth, it is evident on television, school, culture and
examples of government and politicians. This influence is subconscious but
strong and they teach the American that the only thing that one needs to
do is to work faithfully and buy the right things and eventually he will get
the perfect life. Then, in the American mind the definition of event (sic: success) is to be rich, be beautiful and young and have a lot of the best and rich
things.]
Here we see a variety of uses of adjectives, such as nominalized adjectives
(e.g. el americano ‘Americans’), descriptor chains (e.g. subconsciente pero fuerte
‘subconscious but strong’), and the copula ser or estar in predicative sentences.
Furthermore, this sub-discourse type is probably more challenging to produce
since the third-year learners produced more descriptive expository features.
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
The positive cluster of dimension 2—the only significant one—represents a
discourse that is both expository and descriptive. Thus, to the extent that
this cluster represents a sub-discourse of the expository type identified in dimension 1, it seems that these learners do use lexico-grammatical features (e.g.
adjectives with different morphological information) reflecting descriptive
elements in expository writing but not as much as narrative elements. This
cluster also contains measures of lexical complexity: long words and type/
token ratio. Adjectives add informational density to learners’ written prose.
Still, post-nominal adjectives predominate in texts with this cluster.
Additionally, at these instructional levels, a low degree of inflectional accuracy
(i.e. agreement errors) is common.
Texts involving expository prose and description have large average summed
z-scores. Figure 4’s similarity to Figure 3 corroborates that this cluster is a
specialized expository discourse. The difference becomes more evident upon
consideration of the Descriptive expository essay section, which was written by
a third-year learner.
18
ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
The third-year learners averaged a higher summed z-score on this cluster
(M = 2.8; SD = 5.3) than the second-year learners (M = "1.4; SD = 5.0), with
the difference being statistically significant [F(1, 634) = 75.4; p # 0.00].
Learners at these levels exhibit some degree of morphological and syntactic
sophistication when attempting to use lexico-grammatical features such as
adjectives with inflections for number and gender in various morphological
and syntactical (i.e. attributive and predicative roles) functions. Also, long
words tend to be dense in derivational morphology and so, in content; a
high type/token ratio indicates that various adjectives are employed in this
discourse. Finally, the inclusion of several adjectival features reflects a slightly
more developed syntax (Parodi 2007), since word order is a critical consideration with Spanish adjectives, which can be pre- or post-nominal, and descriptors are often chained together (i.e. X has qualities A, B, and C).
The negative cluster of features—singular nouns and the lexical density
measure—does not constitute a text type. Instead, it indicates what descriptive
expository texts lack. Referential lexical items in the form of nouns are missing
as well as a variety of parts of speech. Thus, descriptive expository texts lack
overall specificity and contain numerous and varied adjectives.
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Figure 4: Concentration of descriptive expository prose features by text type
and averaged summed z-score
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
19
Factor 3: Expository prose with a stance
The six positive features of dimension 3 are indicative of discourse where
writers are reporting both their own thoughts about the veracity of eventualities as well as others’ perspectives. This cluster includes mostly lexical features
that indicate writers’ stance (Biber et al. 2006), or their commitment to the
truth of some proposition, taking the form of verbs of knowledge, cognitive
verbs, and verbs of probability (cf. Collentine 1995). The use of a grammatical
feature such as third-person verb inflections is also said to minimize the writer’s risk of misconstruing the reference in the discourse (Castellano 2000). The
cluster also includes complex syntactic features: the subordinating conjunction
que and relative pronouns without prepositions (i.e. que, quien, cual), which
help the reader to identify references in the discourse.
Figure 5’s detailing of the average summed z-scores by text type confirms the
expository function of this cluster.
Argumentative essay. El autor Mario Benedetti posiblemente quiere explicar la imaginación de los niños. Los niños no piensan lógicamente porque no
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
Figure 5: Concentration of expository prose with stance features by text type
and averaged summed z-score
20
ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
Factors 4 and 5
The last two factors did not include enough lexical and grammatical features to
meet the five-feature threshold above the !0.30 level. However, Factors 4 and
5 share some linguistic features. For both factors, some elements in nominal
clauses (i.e. singular and masculine adjectives in Factor 4, and plural adjectives
and definite articles in Factor 5) appeared in complementary distribution with
infinitive verbs. This is probably evidence to support the notion that nominal
and verbal elements appear in complementary distribution in these learners’
writing.
DISCUSSION AND CONCLUSIONS
The development of learners’ interlanguage can be seen as the acquisition of
the ability to create meanings by combining lexical, sentential, pragmatic, and
discourse features. This study describes whether and how Spanish learners use
lexico-grammatical features to produce interlanguage-specific discourse types.
In response to our initial research questions, we found that second- and
third-year Spanish L2 learners’ lexical and grammatical features clustered
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
saben la realidad. Estaba la primera vez que Osvaldo mire la televisión.
Cuando un niño o niña mira una persona mirando y hablando en su dirección, piensan que están hablando a ellos. Es una que nosotros no pensar,
porque nos sabemos la realidad. También, los niños piensan que los personas
en el televisor son sus amigos. [The author Mario Benedetti possibly wants to
explain the imagination of children. Children do not think logically because
they do not know reality. It was the first time Osvaldo watched TV. When a
boy or a girl watches a person looking at or talking in their direction, they
think they are talking to them. It is one that we do not to think (sic: think
about), because we know the reality.]
The writer here speculates about an author’s intention and argues for a
specific motive by reporting on the thoughts and attitudes of the characters
in the story.
Dimension 3’s negative cluster only indicates what is missing where stance
predominates. The features include lexical and grammatical narrative tools,
namely, imperfect verbs, progressive aspect, and prepositions. These features
describe the scene or the background context (e.g. time, place, people, and
other co-occurring events) in which the main events occur, suggesting perhaps
that these learners do not couple stances with narrative elements. Support for
this is found in a comparison of the two levels of learners. The data indicate
that expository prose with a stance occurs mostly in less proficient learners,
which would explain why it is disassociated with certain narrative elements.
The second-year learners averaged a higher summed z-score on this cluster
(M = 3.9; SD = 4.5) than the third-year learners (M = 2.2; SD = 2.7), with the
difference being statistically significant [F(1, 634) = 17.6; p # 0.00].
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
21
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
reliably into five factors that mainly combined verbal or nominal features. The
analysis uncovered four significant clusters that can be considered distinct
discourse types. The communicative functions of these discourse types break
down into two main stylistic variations: narrative (characterized by verbal
features) and expository (characterized by nominal features).
Only 20% of the variance in learners’ interlanguage could be explained by
the five factors while Biber et al.’s (2006) native-speaker multidimensional
analysis accounted for 45% of the targeted corpus’ variance. Biber et al.’s
(2006) clusters were more numerous and contained more features. And, although their native-speaker analysis included data from the written and oral
modes, whereas this study included data from only the written mode, the L2
clusters possibly had few features because clustering of features of any kind
develops slowly in the L2, and the psycholinguistic consolidation of productive
associations requires time and practice (Collentine and Asención-Delaney
2010). If so, it should also not be surprising that the discourse types favored
by Spanish learners at the second and third year of university-level instruction
are novel in comparison with native-speaker clusters in a number of respects.
Additionally, the fact that the clusters are relatively homogeneous in terms of
the parts of speech they represent (i.e. either verbal or nominal) indicates that
these learners depend on a limited repertoire of lexico-grammatical features to
achieve their communicative goals.
The L2 expository dimension constitutes a concentrated use of nominal features. This discourse type represents a certain discourse complexity since in
Biber’s (2001) study of English texts and in Parodi’s (2007) study of Spanish
texts, nominal features work together to generate semantically dense discourse
that is informationally rich. This is corroborated by the observation that
advanced-low third-year learners produced more expository discourse.
However, the learner discourse type uncovered here differs from
native-speaker expository discourse in that intermediate-high second-year
learners produce narrative elements in argumentation, such as arguing for a
point with an illustrative story. Future research may reveal whether the narrative elements constitute a compensatory strategy as reported by Hinkel
(2004) with English learners and/or a function of the Spanish curriculum,
which highly emphasizes verb morphology (i.e. conjugations for preterit and
imperfect).
The narrative dimension showed that Spanish learners’ narrative discourse
is not limited to presenting an account of events in the past, but it also reflects
learners’ personal speculations, feelings, and attitudes towards the events. This
involvement may be due to task requirements or to beginner–intermediate
learners’ tendency to produce the L2 to talk about themselves, which is typical
in the Spanish L2 curriculum.
Finally, the remaining two discourse types are probably more or less useful
to L2 Spanish writers, depending on their level of development. The descriptive expository prose is probably a more sophisticated version of learner expository prose in general, since the advanced-low learners utilized it more. On
22
ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
NOTES
1 Morphological diversity occurs when a
learner uses a variety rather than a
small subset of the language’s inflectional and derivational morphemes
(Howard 2002, 2006). Morphological
complexity occurs where learners
either employ words with multiple derivational morphemes or marked inflectional morphemes (e.g. subjunctive,
conditional, future, past participle).
2 Just as digital technologies have altered
various classroom practices, so has the
ability to amass large corpora of digitized representations of the L2 changed
how we design pedagogical grammars
(O’Keeffe et al. 2007). Corpus-based
data-driven learning strategies are beginning to be incorporated into pedagogical materials (Boulton 2009).
3 A lexical category is one identified by a
search expression or measurement (e.g.
lexical density ratio) focusing on the
lexical properties of a word or segment,
which primarily entails checking a
word’s lemma. A grammatical category
is one whose search expression
focuses on the inflectional properties
of a word or segment (e.g. person,
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
the other hand, the expository prose with a stance is peculiar in that it contains
features associated with stance and subordination, and yet the second-year
learners used it more than the third-year learners. It was, nevertheless,
shown that this discourse type does not appear intermixed with other important feature sets, as it is disassociated with important narrative features. The
second-year learners may only have been able to produce segments with these
features and not simultaneously generate other discourse types. Clearly, the
emergence of stance and how it is operationalized in L2 writing deserves a
more fine-grained analysis in future research.
The findings provide examples of the multiple ways that linguistic complexity occurs in the L2. Although Spanish learners’ discourse did not show signs of
syntactic complexity (e.g. frequent use of relative clauses, subordinate clauses,
use of clitics; cf. Wolfe-Quintero et al. 1998; Ortega 2000), the frequent use of
nominal features affects informational density due to the presence of numerous derivational morphemes. Inflectional complexity in the form of marked
forms was not predominant in the data set. Still, the learners’ verbal inflections
did vary, which is a sign of L2 development occurring (Howard 2002, 2006;
Collentine 2004; Marsden and David 2008).
Regarding the study’s limitations, the multidimensional analysis would have
benefited from the inclusion of spontaneous learner samples (e.g. emails, chat
scripts) to understand L2 discourse where there are high online cognitive demands. Additionally, our data are limited to intermediate-high and
advanced-low learners of Spanish. Finally, given the nature of the multidimensional analysis that examines what appears in communication, as opposed
to what does not, the present analysis does not consider learner errors (Grant
and Ginther 2000). Nevertheless, the data and the qualitative analyses provide
complementary perspectives on how L2 communication occurs in relatively
extended discourse where learners must consider numerous lexical, inflectional, derivational, and syntactic features.
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
number, tense); it may also entail
high-frequency functors whose role is
largely syntactic (e.g. si ‘if’, articles,
comparative constructions). A lexicogrammatical category is identified
by both lexical and grammatical properties, such as feminine nouns, and adverbial conjunctions.
4 Lexical density is derived from the
total number of non-functor words
23
(nouns + adjectives + verbs + derived
adverbs) divided by the total words in a
text.
5 Long words are defined here as any
word whose length is greater than the
mean word length of the corpus plus
the standard deviation word length of
the corpus, e.g. mean word length:
4.6 + SD: 4.9 = 10.5–11.
American Council on the Teaching of
Foreign
Languages.
2001.
‘ACTFL
Proficiency Guidelines – Writing, available at
http://www.actfl.org/i4a/pages/index.cfm?
pageid=3326. Accessed 27 September 2010.
Bachman, L. 1990. Fundamental Considerations
in Language Testing. Oxford University Press.
Belz, J. 2004. ‘Learner corpus analysis and the
development of foreign language proficiency,’
System 32/4: 577–97.
Biber, D. 1988. Variation across Speech and
Writing. Cambridge University Press.
Biber, D. 2001. ‘On the complexity of discourse
complexity: A multidimensional analysis’
in S. Conrad and D. Biber (eds): Variation in
English: Multidimensional Studies. Longman,
pp. 215–40.
Biber, D. and S. Conrad. 2001. ‘Introduction:
Multidimensional analysis and the study of
register variation’ in S. Conrad and D. Biber
(eds): Variation in English: Multidimensional
Studies. Longman, pp. 3–13.
Biber, D., M. Davies, J. Jones, and N. TracyVentura. 2006. ‘Spoken and written register
variation in Spanish: A multi-dimensional analysis,’ Corpora 1: 1–37.
Boulton, A. 2009. ‘Testing the limits of
data-driven learning: Language proficiency
and training,’ ReCALL 21/1: 37–54.
British
National
Consortium.
2001.
‘BNC World [Database], available at http://
www.natcorp.ox.ac.uk/. Accessed 09 July
2009.
Canale, M. and M. Swain. 1980. ‘Theoretical
bases of communicative approaches to second
language teaching and testing,’ Applied
Linguistics 1/1: 1–47.
Castellano, A. 2000. ‘Ambigüedad y variación
del pronombre personal sujeto’ in M. Muñoz,
G. Fernández, A. Rodrı́guez, and V. Benı́tez
(eds): IV Congreso de Lingüı́stica General.
Universidad de Cádiz, pp. 521–31.
Cheng, C., H-C. Lu, and P. Giannakouros.
2008. ‘The uses of Spanish copulas by
Chinese-speaking learners in a free writing
task,’ Bilingualism: Language and Cognition 11/
3: 301–17.
Collentine, J. 1995. ‘The development of complex syntax and mood-selection abilities by
intermediate-level
learners
of
Spanish,’
Hispania 78/1: 123–36.
Collentine, J. 2004. ‘The effects of learning contexts on morphosyntactic and lexical development,’ Studies in Second Language Acquisition 26/
2: 227–48.
Collentine, J. 2009. ‘Study abroad research:
Findings, implications and future directions’
in M. Long and C. Doughty (eds): The
Handbook of Language Teaching. WileyBlackwell Press, pp. 218–34.
Collentine, J. and Y. Asención-Delaney. 2010.
‘A corpus-based analysis of the discourse functions of Ser/Estar+adjective in three levels of
Spanish FL learners,’ Language Learning 60/2:
409–45.
Ellis, N. C. 2002. ‘Reflections on frequency effects in language processing,’ Studies in Second
Language Acquisition 24/2: 297–339.
Ellis, N. C. 2006. ‘Language acquisition as rational contingency learning,’ Applied Linguistics
27/1: 1–24.
Ellis, N. C., R. Simpson-Vlach, and C. Maynard.
2008. ‘Formulaic language in native and second
language speakers: Psycholinguistics, corpus
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
REFERENCES
24
ANALYSIS OF A WRITTEN L2 SPANISH CORPUS
Marsden, E. and A. David. 2008. ‘Vocabulary
use during conversation: a cross-sectional
study of development from year 9 to year 13
among learners of Spanish and French,’
Language Learning Journal 36/2: 181–98.
Myles, F. 2005. ‘Interlanguage corpora and
second language acquisition research,’ Second
Language Research 21/4: 373–91.
Myles, F. and R. Mitchell. 2005. ‘Using information technology to support empirical SLA
research,’ Journal of Applied Linguistics 1/2:
169–96.
O’Keeffe, A., M. McCarthy, and R. Carter.
2007. From Corpus to Classroom. Cambridge
University Press.
Ortega, L. 2000. Understanding Syntactic
Complexity: The Measurement of Change in the
Syntax of Instructed L2 Spanish Learners.
Unpublished dissertation, University of Hawaii
at Manoa.
Parodi, G. 2007. ‘Variation across registers in
Spanish: Exploring the El Grial PUCV Corpus’
in G. Parodi (ed.): Working with Spanish
Corpora. Continuum, pp. 11–53.
Skiba, R. and N. Dittmar. 1992. ‘Pragmatic, semantic, and syntactic constraints and grammaticalization: A longitudinal perspective,’
Studies in Second Language Acquisition 14/3:
323–49.
Upton, T. A. and U. Connor. 2001. ‘Using computerized corpus analysis to investigate the
textlinguistic discourse moves of a genre,’
English for Specific Purposes 20/4: 313–29.
Weber, E. and P. Bentivoglio. 1991. ‘Verbs of
cognition in spoken Spanish: A discourse profile’ in S. Fleischman and L. Waugh (eds):
Discourse Pragmatics and the Verb: Evidence from
Romance. Routledge, pp. 194–213.
Wolfe-Quintero, K., S. Inagaki, and S. Kim.
1998. ‘Second language development in
writing: measures of fluency, accuracy and
complexity’,
Technical
Report
#
17.
University of Hawai’i, Second Language
Teaching and Curriculum Center.
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011
linguistics, and TESOL,’ TESOL Quarterly 42/3:
375–96.
Forsberg,
F.
2005.
‘Ready-to-speak’:
Prefabricated sequences in spoken French as
a second language and as a native language.
[‘Prêt-à-parler’ séquences préfabriquées en français
parlé L2 et L1],’ Cahiers De l’Institut De
Linguistique De Louvain 31: 183–95.
Geyer, N. 2007. ‘Self-qualification in L2
Japanese: An interface of pragmatic, grammatical, and discourse competences,’ Language
Learning 57/3: 337–67.
Givón, T. 1985. ‘Function, structure, and language acquisition’ in D. Slobin (ed.): The Cross
Linguistic Study of Language Acquisition.
Lawrence Erlbaum, pp. 1008–25.
Granger, S., J. Hung, and S. Petch-Tyson.
2002. Computer Learner Corpora, Second
Language Acquisition and Foreign Language
Teaching. John Benjamins.
Grant, L. and A. Ginther. 2000. ‘Using
computer-tagged linguistic features to describe
L2 writing differences,’ Journal of Second
Language Writing 9/2: 123–45.
Hinkel, E. 2004. ‘Tense, aspect and the passive
voice in L1 and L2 academic texts,’ Language
Teaching Research 8/1: 5–29.
Horn, L. and G. Ward. 2006. The Handbook of
Pragmatics. Wiley-Blackwell.
Howard,
M.
2002.
‘Prototypical
and
non-prototypical marking in the advanced
learner’s aspectuo-temporal system,’ EuroSLA
Yearbook 2: 87–114.
Howard, M. 2006. ‘Variation in advanced
French interlanguage: A comparison of three
(socio)linguistic variables,’ Canadian Modern
Language Review 62/3: 379–400.
Kasper, G. 2001. ‘Four perspectives on L2 pragmatic development,’ Applied Linguistics 22/4:
502–30.
Klein, W. and C. Perdue. 1997. ‘The basic variety (Or: Couldn’t natural languages be much
simpler?),’ Second Language Research 13/4:
301–47.
Longacre, R. 1983. The Grammar of Discourse.
Plenum.
Y. ASENCIÓN-DELANEY AND J. COLLENTINE
25
NOTES ON CONTRIBUTORS
Yuly Asención-Delaney is an Associate Professor of Spanish at Northern Arizona
University. Her research interests include corpus linguistics, Spanish second language
acquisition, second language writing, and second language assessment. Address for correspondence: Northern Arizona University, Modern Languages, Box 6004, Flagstaff, AZ
86011, USA. <[email protected]>
Joseph Collentine is Professor of Spanish and Chair of the Modern Languages Department
at Northern Arizona University. His research interests include the study of L2 morphosyntax, the acquisition of mood, corpus linguistics, study abroad, and computer assisted
language learning.
Downloaded from applij.oxfordjournals.org by guest on January 29, 2011