Content uploaded by Laia Cutillas i Alberich
Author content
All content in this area was uploaded by Laia Cutillas i Alberich on Jan 21, 2017
Content may be subject to copyright.
LAIA CUTILLAS, LILIANA TOLCHINSKY, ELISA ROSADO, JOAN
PERERA
Indicators of lexical growth throughout age, genre
and modality for a Catalan L1 corpus
Abstract
Lexical development is a key facet of later, school-age language
development. The goal of the study is twofold: on the one hand, to
describe quantitatively the text-embedded lexicon of a corpus of texts
produced by informants from nine years of age to adulthood, and on
the other hand, to identify which characteristics of the lexicon can be
considered as indicators of age, discourse genre, and modality of
production differentiation.
The GRERLI-CAT1 (Grup de Recerca per a l’Estudi del
Repertori Lingüístic, Català L1) corpus is constituted by a set of
narrative and expository spoken and written texts in Catalan that were
produced by 79 bilingual Catalan/Spanish speakers that have Catalan as
their home language. They were distributed in five groups according to
level of education: elementary, secondary, high school and university
level, and language teachers. The corpus comprises 316 texts, which
include a total of 84,081 tokens, 40,612 types and 31,811 lemmas.
Four measures were applied for characterizing the corpus
lexically: (1) lexical diversity, (2) lexical density, (3) word length, and
(4) productivity of verbs. Lexical diversity was selected to gauge the
range of vocabulary displayed in the texts, lexical density as an
indicator of textual richness and informativeness, word length was
taken as an indicator of lexical complexity, and productivity of verbs
as an indicator of lexical richness within the verbal domain. We
discuss the utility of these four measures as indicators of text
construction development in different genres and modalities.
160 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
1. Introduction
The general framework of this study is Corpus Linguistics, which
approaches the study of language empirically. It is based on authentic
samples of language use, so data treatment is external to the speaker,
directly observable and, therefore, demonstrable. It uses computerized
corpora, which represent an important source of quantitative
information. The use of these samples enables to determine frequency
of occurrences and lack of specific linguistic elements (Civit 2003: 2).
Corpus linguistics is a method applicable to every area of linguistic
research, from phonology to discourse. Moreover, corpus-based
research is applicable to linguistic education both in first (L1) and
second (L2) language learning. Native corpora, like the one used in
this study, can be very useful for education professionals because they
show what speakers say or write in a specific communication
situation, as well as their typical difficulties (Nesselhauf 2004: 144-
145). In this study, this methodology is applied to first language
acquisition and, more specifically, to the later, school age language
development of Catalan. Catalan is a Romance language that typically
displays a rich morphological system: nouns and adjectives are
inflected for gender and number; pronouns are inflected for gender,
number, person and case; and verbs are inflected for person/number,
tense/aspect and mood. Moreover, it requires gender and number
agreement between noun and adjective and between pronoun and its
antecedent, as well as person and number agreement between subject
and verb.1
1 Catalan is spoken in four Spanish autonomous communities (Aragon, the
Balearic Islands, Catalonia and the Valencian Community), in the French re-
gion of Rousillon, and in the city of Alghero, on the Italian island of Sardinia.
In Catalonia, Catalan and Spanish are equally recognized as official language,
but Catalan is the language of education. Therefore, children educated in
Catalonia are bilingual (Catalan/Spanish) living in a multilingual environ-
ment: “Due to a major surge of immigration over the past decade (3% in 2000
to 13% in 2008), an increasing percentage of children speak a language differ-
ent from Catalan and Spanish at home” (Llauradó / Tolchinsky in press).
Indicators of lexical growth for a Catalan L1 corpus 161
Several Catalan corpora have been compiled during the recent
years. The CCCUB (Corpus del Català Contemporani de la UB)
(Boix et al. 2002), compiled by the Grup d’Estudi de la Variació
(GEV) at the Universitat de Barcelona (UB), contains texts in
different dialects and registers, but only in one modality of production,
the spoken one. The AnCora-CA (Anotated Corpus-Catalan) (Taulé et
al. 2008) includes written journalistic texts, so it could only be
assessed one genre and one modality. The CesCa (Català Escolar
Escrit a Catalunya) (Llauradó et al. 2012) contains different types of
written texts: narrations of a film storyline, recommendations of a
film, or definitions of words and jokes, but only in one modality of
production. These two latter corpora are compiled by the Centre de
Llenguatge i Computació (CLiC) of the UB. The CICA (Corpus
Informatitzat del Català Antic) (Torruella 2009), compiled by the
Grup de Lexicografia i Diacronia (SFI) of the Universitat Autònoma
de Barcelona (UAB), is a diachronic corpus which includes books
written from the 12th to the 15th centuries. The CTILC (Corpus
Textual Informatitzat de la Llengua Catalana) (Rafel / Solanellas
1986), compiled by the Institut d’Estudis Catalans (IEC) with
lexicographic purposes, contains both literary and non- literary texts in
different genres but, again, only in the written modality. The
GRERLI-CAT1 corpus, compiled by the GRERLI (Grup de Recerca
per a l’Estudi del Repertori Lingüístic) at the UB, includes texts in
different genres (narrative and expository) and modalities of
production (spoken and written), produced by subjects of different
school levels (elementary, secondary and high school, university level
and language teachers).
The goals of this paper are (1) to analyse quantitatively the
GRERLI-CAT1 corpus text-embedded lexicon and (2) to identify
which characteristics of the lexicon can be considered as indicators of
lexical growth throughout age, genre, and modality. With this
purpose, four measures are evaluated: lexical diversity, lexical
density, word length and productivity of verbs.
Lexical diversity (Malvern et al. 2004: 19) is selected because it
can gauge the range of vocabulary use. Lexical density (Read 2000:
200) is considered a good indicator of textual richness and
informativeness. Word length (Strömqvist et al. 2002: 48), is chosen
162 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
as a measure of lexical complexity. Productivity of verbs is selected as
an indicator of lexical richness within the verbal domain. This
measure embraces both the diversity of verbal lemmas, that is, the
number of different verbal lemmas used in the corpus, and the
productive use of inflectional morphology of verbs, that is, the number
of different tense forms (Serrat et al. 2004: 221).
2. Corpus compilation
The GRERLI-CAT1 corpus was compiled within the framework of a
cross-linguistic project designed to analyse the development of text
construction abilities in different languages. The corpus was compiled
in 1998 as part of the international project “Developing literacy in
different contexts and in different languages”, Spencer Foundation,
Chicago, United States of America (P.I.: R. A. Berman, Tel Aviv
University). The languages that took part in this study were Dutch, US
English, French, Hebrew, Icelandic, Spanish, and Swedish, with the
subsequent addition of Catalan. The main goals of this project were to
understand how speakers/writers develop their discursive abilities
throughout different educational levels, to analyse how they use the
resources of their respective languages to construct discourse in
different genres (expository and narrative) and modalities of
production (spoken and written), and, finally, to find out common and
language-specific patterns of language use when producing spoken
and written texts (Berman 2002; Aparici 2010).
Language usage, both spoken and written, is always framed
within a specific discourse genre. As Tolchinsky (2004: 235) says:
There is no such thing as neutral use of language: people constantly attune
their speech to specific intentions, purposes, and interlocutors. Thus,
development is viewed in terms of the acquisition of different discourse
genres, and the way that the cultural conventions of genres constrain the use
of linguistic forms.
Indicators of lexical growth for a Catalan L1 corpus 163
The two genres selected for the project were expository and narrative.
Several reasons lead to the selection of these two genres. Firstly,
because the narrative genre has been vastly studied from a
developmental psycholinguistic point of view while the expository
genre has not, since it is a late developing genre, and secondly,
because of the contrast in personal involvement that characterises
these two genres. A first person narrative and the discussion of a topic
illustrate the extremes of a continuum of personal involvement.
Personal involvement of the speaker/writer is demonstrated, among
other things, in the election of the main characters. In personal
narratives the main character is the speaker/writer and, in expository
texts, the topic gains prominence (Aparici 2010).
Regarding modalities of production, spoken and written texts,
differ from the point of view of the demands imposed by processing.
The spoken modality is constrained by on-line processing and can
cause mistakes, breaks, repetitions, reformulations or lack of referents.
On the other hand, the written modality is not subject to time
limitations. It enables the writer to plan thoroughly what he or she
wants to write, to outline or to proofread. For this reason, written texts
are prone to be better organized and be more cohesive and coherent
than spoken texts.
Informants were selected from nine years of age onwards
because at this age they have already attained a command of narrative
structure and of the mechanisms of discursive cohesion (Berman /
Slobin 1994). The sample also included adults because the
development of discursive abilities is a long process that spans from
childhood to beyond adolescence (Berman / Verhoeven 2002: 14).
2.1. Informants
The texts were produced by 79 informants from Barcelona, who were
bilingual speakers of Catalan and Spanish and whose home language
164 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
is Catalan.2 Participants were distributed in five groups, according to
their educational level and age. The elementary school group
comprises 20 students from the 4th year elementary school. The
secondary school group comprises 19 students of the 1st year
secondary school. The high school group comprises 20 students of the
1st year high school. The university group comprises 10 students from
different courses and specialities, specifically, four Science students
and six Humanities students. Finally, the teachers group comprises 10
high school Catalan language teachers. This group is expected to
represent a level of professional use of the language.
Table 1 shows data of all informants’ groups:
Elementary
School
Secondary
School
High School
University
Teachers
n = 20
n = 19
n = 20
n = 10
n = 10
M
Range
M
Range
M
Range
M
Range
M
Range
10;3.
9;6. -
10;9.
13;0.
12;4. -
13;4.
17;4.
16;3. -
18;10.
21;9.
19;1. -
24;2.
43;10.
37;8. -
54;11.
Table 1. Number of informants, mean age and age range.
2.2. Tasks
All informants produced four texts. After watching a three-minute
video without text, participants were asked to produce a spoken
expository text, a written expository text, a spoken narrative text and a
written narrative text. The video shows different conflictive situations
in schools, such as fights, shunning classmates, cheating in exams, etc.
The purpose was that all texts produced by the informants had a
common topic to enable the comparison of the text linguistic
characteristics.
2 It is unlikely to find monolingual speakers of Catalan, since Catalan and Span-
ish are both official languages in Catalonia. All children use Catalan at school
and Spanish is massively present both in the media and in social settings.
Indicators of lexical growth for a Catalan L1 corpus 165
2.3. Procedure
Informants produced the four tasks individually. Data were elicited in
two sessions and four different orders of text production were
followed: A (first session: spoken narrative/written narrative; second
session: spoken expository/written expository); B (first session:
written narrative/spoken narrative; second session: written
expository/spoken expository); C (first session: spoken
expository/written expository; second session: spoken
narrative/written narrative); and D (first session: written
expository/spoken expository; second session: written
narrative/spoken narrative).
2.4. Corpus storage
We have used two different versions of the corpus: clean (net) and
morphologically tagged (morfo). In the following sub-sections, the
characteristics of these two corpus versions will be described.
2.4.1. Clean version (net)
This version consists of text files without format which contain the
production with no additional symbols or indications of spelling
mistakes. All marking of ancillary material, such as repetitions,
reformulations, pauses and comments, is omitted. The transcription
unit is the clause. That is, each text line in this version corresponds to
a clause. We follow Berman and Slobin’s (1994: 660) definition of
this unit:
We define a clause as any unit that contains a unified predicate. By unified,
we mean a predicate that expresses a single situation (activity, event, state).
Predicates include finite and non-finite verbs, as well as predicate adjectives.
This version aims at producing an input file for the morphological
analyser HSMorfo, which performs the morphological tagging. The
counts of tokens and types were computed using the net version.
166 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
2.4.2. Morphologically-tagged version (morfo)
This is a text version that results from the net version of the corpus
after having been morphologically tagged using the HSMorfo Linux
software.3 Files of the morphologically analysed texts contain three
columns: the first one represents the type, the second one the lemma
and the third one includes the EAGLES (Expert Advisory Group on
Language Engineering Standards) tag (EAGLES 1996), which shows
the grammatical category of the type. The measures for the four
analysis dimensions of the current paper were calculated using the
morfo version.
2.4.3. Other versions
There are three more versions of the corpus: mirror (rep), normalized
(nor) and CHAT (cha). The mirror version (only for writing texts)
consists in the transcription of the written productions into a MS Word
file, with the original disposition of paragraphs and lines, punctuation
and spelling. In the normalized version, texts (both spoken and
written) are transcribed in CHAT (Codes for the Human Analysis and
Transcripts) format (MacWhinney 2012). Spoken productions are
transcribed orthographically (not phonetically) including processing
information (pauses, repetitions, reformulations, etc.). In written texts,
following CHAT conventions, spelling mistakes are followed by the
correct word (e.g. vastant [: bastant] ‘enough’). Finally, the CHAT
version only differs from the nor version in that the transcription unit
is the clause.
3 It is important to remark that some files have been analysed using the
FreeLing software, because they included a character which provoked a mal-
functioning on HSMorfo and it finished the analysis before arriving to the
end. However, the working method of both programs is the same, because
they use the same analysis process, as well as EAGLES tags on the tagging
process. They only differ in the treatment of the apostrophized words
(FreeLing attaches them to the next word and HSMorfo separates them) and in
the contractions (FreeLing separates them into two words and HSMorfo treats
them as one word). These treatments have been subsequently fixed using a
software specially designed to correct these errors.
Indicators of lexical growth for a Catalan L1 corpus 167
2.5. Corpus processing
The net version provides the input files for the HSMorfo Linux
software, which performs the morphological analysis and tagging, the
morfo version of the corpus. The HSMorfo software uses the tags
developed by the EAGLES group for morphosyntactic annotation of
European languages, which are adapted to Catalan in Civit (2003).
Each position of the tag corresponds to a feature (whose number
varies for each category), the value of this feature and a code
representing them. Zero shows that a given feature has no value.
Below are some examples of the morphological analysis of three
specific cases of segmentation, which could be a problem for
morphological analysers:
• Multi word expressions, linked by an underscore (_), such as punt
de vista ‘point of view’:
punt de vista
1. punt_de_vista punt_de_vista (‘point of view’) NCMS000
(Noun Common Masculine Singular)
• Words containing a dash (-) and/or an apostrophe (’):
a) Words with clitic pronouns, which are attached to a verb
using a dash and/or an apostrophe, such as controla’ls-ho
‘control it to them’:
controla’ls-ho
1. controla controlar (‘to control’) VMM02S0 (Verb Main
Imperative 2nd Singular)
2. ’ls ell (‘he’) PP3CP000 (Personal Pronoun 3rd Common
Plural)
3. -ho ho (‘it’) PP3NN000 (Personal Pronoun 3rd Neuter
Invariable)
b) Apostrophized words, such as l’únic ‘the unique’:
l’únic
1. l’ el (‘the’) DA0CS0 (Determinate Article Common
Singular)
2. únic únic (‘unique’) AQ0MS0 (Adjective Qualifying
Masculine Singular)
168 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
• Contractions (prepositions attached to an article) are treated by the
analyser as compound prepositions, and they include grammatical
information about gender and number. In other words, they
present the preposition features (three first values of the tag) and
the article features (the two last values of the tag), such as dels ‘of
the’:
dels
1. dels del (‘of the’) SPCMP (Adposition Preposition
Complex Masculine Plural)
As we have seen in this section, the characteristics of the GRERLI-
CAT1 corpus are as follows: it contains 316 texts, which include a
total of 84,081 tokens, 40,612 types and 31,811 lemmas. The main
features that identify this corpus are: (i) Language: the corpus contains
texts in Catalan; (ii) Developmental dimension: the corpus reflects
later, school age language development, because it contains data
produced by informants from nine years old to adulthood; (iii) Genre:
it contains texts in two discourse genres: narrative and expository; (iv)
Modality: it includes texts of two modalities of production: spoken
and written; and finally, (v) Corpus storage: the corpus is stored in
five different versions that can serve as input for different
computational platforms. Moreover, because the corpus is part of a
cross-linguistic project, in which the same data elicitation procedure
was used for the different languages, it can be compared with the
corpora in the other languages that took part in the study.
3. Dimensions of analysis
In order to identify the characteristics of the lexicon that can be
considered as indicators of age, discourse genre and modality of
production differentiation, the selected measures are: (3.1.) lexical
diversity, calculated using type-token, lemma-token and lemma-type
ratios; (3.2.) lexical density, i.e. the proportion of content words
Indicators of lexical growth for a Catalan L1 corpus 169
relative to the total number of words; (3.3.) word length, i.e. the
number of letters of each lexical word; and (3.4.) productivity of
verbs, characterised by the diversity of verbal lemmas used and the
productive use of inflectional morphology of verbs, i.e. the number of
verb tenses associated to each verbal lemma.
3.1. Lexical diversity
Lexical diversity (Malvern et al. 2004) is a widely used measure of
language development, because it can gauge the range of vocabulary
use. Although the type-token ratio is by far the most commonly used
measure of lexical diversity, it is much less useful in measuring
vocabulary of inflectional languages like Catalan than another
measure, the lemma-token ratio, which provides more accurate
information about lexical diversity, as Granger / Wynne (2000: 251)
say:
A learner who uses five different forms of the verb go (go/goes/going/
gone/went) in one and the same text has a less varied vocabulary than the one
who uses five different lemmas (such as go/come/leave/enter/return).
Moreover, a parameter that calculates the correlation between lemmas
and types, the lemma-type ratio, can also be useful to study lexical
diversity of Catalan. The type-token ratio is obtained by dividing the
number of types by the number of tokens, the lemma-token ratio by
dividing the number of lemmas by the number of tokens; and finally,
the lemma-type ratio by dividing the number of lemmas by the
number of types.
3.2. Lexical density
Lexical density (Read 2000) is considered to be a good indicator of
textual richness. If content words convey the vast bulk of semantic
content, then the relative proportion of lexical words used in a text can
provide an idea of its informativeness. Lexical density is obtained by
170 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
dividing the number of lexical words (nouns, main verbs, adjectives
and adverbs) by the total of words in a sample.
3.3. Word length
Word length (Strömqvist et al. 2002) is regarded as an indicator of
lexical complexity. The longer words are supposed to be
derivationally more complex words, so word length is a suitable
measure of lexical complexity. Function words (auxiliary verbs,
determinants, pronouns, prepositions, conjunctions and interjections)
have been removed from this count because of their grammatical
status. Means for word length are obtained by counting the number of
letters in each lexical word over the total number of words.
3.4. Productivity of verbs
Productivity of verbs is seen here as an indicator of lexical richness
within the verbal domain. In the context of the present study we
consider the diversity of verbal lemmas (number of different lexical
verbs that appear in the corpus) apart from the productive use of
the inflectional verb morphology (number of different verb forms –
marking of tense/aspect or mood– for each specific verbal lemma).
Diversity of verbal lemmas is obtained by dividing the total of verbal
lemmas by the number of different verbal lemmas used in a sample.
Productive use of inflectional morphology (Serrat et al. 2004) is
calculated by counting the number of different tenses used for each
verbal lemma. For these counts, differences in person/number
inflection have not been considered, and the non-finite forms of the
verb (infinitive, participle and gerund) have only been considered as
part of a compound verb. That is, tenses formed by an auxiliary verb
(haver ‘to have’ or anar ‘to go’, and ser ‘to be’, for passive forms)
plus a non-finite form of the main verb.
Indicators of lexical growth for a Catalan L1 corpus 171
4. Results
4.1. Quantitative description of the corpus
Table 2 presents the distribution of tokens, types and lemmas by text
type. There are some differences between both the two discourse
genres and the two modalities of production. Expository texts are
significantly longer than narrative texts, both spoken and written: (F
(1, 74) = 19.024, p = .000) for tokens, (F (1, 74) = 22.661, p = .000)
types, (F (1, 74) = 24.078, p = .000) and lemmas, respectively. Spoken
texts in both genres have significantly more tokens, types and lemmas
than written texts: (F (1, 74) = 25.669, p = .000) for tokens, (F (1, 74)
= 12.080, p = .001) types, (F (1, 74) = 12.556, p = .001) and lemmas,
respectively. By text types, spoken expository has more tokens than
spoken narrative and written expository than written narrative. Types
and lemmas present a different pattern: expository texts, both spoken
and written, have more types and lemmas than narrative ones. A
significant interaction between genre and modality is also found for
tokens (F (1, 74) = 11.423, p = .001), types (F (1, 74) = 11.113, p =
.001) and lemmas (F (1, 74) = 8.451, p = .005).
Table 3 presents the distribution of tokens, types and lemmas by
age group. There is a significant effect of age: (F (4, 74) = 13.910, p =
.000) for tokens, (F (4, 74) = 24.176, p = .000) types, (F (4, 74) =
24.047, p = .000) and lemmas, respectively. High school group has the
highest number of tokens, types and lemmas, followed by the older
groups, university and teachers. The next group is secondary school
and elementary school is the age group with the lowest number of
tokens, types and lemmas. So the counts of tokens, types and lemmas
increases gradually from elementary to high school, and then
decreases progressively from high school to university and then from
university to teachers. However, the means of tokens, types and
lemmas show a different pattern: there is an increase from elementary
school to university, but then the means of tokens, types and lemmas
decrease in the teachers’ group. An interaction is found between genre
and age for tokens (F (4, 74) = 4.660, p = .002), types (F (4, 74) =
172 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
3.748, p = .008) and lemmas (F (4, 74) = 3.732, p = .008). The
number of tokens also shows an interaction between modality and age
(F (4, 74) = 2.574, p = .045).
Counts
M
SD
Spoken
Expository
Tokens
30,315
383.73
406.69
Types
12,269
155.30
107.77
Lemmas
9,479
119.99
79.64
Written
Expository
Tokens
16,929
214.29
143.39
Types
9,828
124.41
65.19
Lemmas
7,791
98.62
51.38
Spoken
Narrative
Tokens
22,151
280.39
275.56
Types
9,730
123.16
77.97
Lemmas
7,658
96.94
59.63
Written
Narrative
Tokens
14,686
185.90
104.56
Types
8,785
111.20
53.98
Lemmas
6,883
87.13
41.98
Table 2. Tokens, types and lemmas distribution by text type.
Counts
M
SD
Elementary
School
Tokens
11,223
140.28
57.63
Types
6,356
79.45
25.62
Lemmas
5,013
62.66
18.63
Secondary
School
Tokens
12,200
160.52
71.70
Types
6,592
86.73
28.19
Lemmas
5,137
67.59
19.55
High School
Tokens
26,082
326.02
192.00
Types
12,059
150.73
56.26
Lemmas
9,252
115.65
41.66
University
Tokens
19,627
490.67
282.00
Types
8,241
206.02
69.98
Lemmas
6,460
161.50
53.38
Teachers
Tokens
14,949
373.72
75.37
Types
7,364
184.10
30.50
Lemmas
5,949
148.72
25.90
Table 3. Tokens, types and lemmas distribution by age.
Indicators of lexical growth for a Catalan L1 corpus 173
4.2. Lexical diversity
Figure 1 illustrates the results by text type of the three measures
selected for characterising lexical diversity: types per token, lemmas
per token and lemmas per type ratios. Firstly, the results of the type-
token ratio are presented. Regarding the differences between the two
discourse genres, no significant effect of genre is found. As for
modality, type-token ratio is significantly higher in the written than in
the spoken texts (F (1, 74) = 175.278, p = .000). By text types, written
texts, both expository and narrative, have the highest type-token ratio,
followed by spoken narrative and spoken expository. There is an
almost significant interaction between genre and modality (F (1, 74) =
3.568, p = .063). Secondly, the lemma-token ratio results show that,
like in the type-token ratio, genre has no significant effect. The
lemma-token ratio is significantly higher in the written than in the
spoken modality (F (1, 74) = 100.689, p = .000). By text types, the
lemma-token ratio is higher in spoken narrative texts than in spoken
expository, but in written narrative is lower than in written expository.
This ratio shows an interaction between discourse genre and modality
of production (F (1, 74) = 4.349, p = .040). Finally, the results for the
lemma-type ratio are presented. This ratio is not affected by genre or
modality. Spoken narrative texts have the highest lemma-type ratio,
followed by written expository, spoken expository and written
narrative. An almost significant interaction between genre and
modality is found (F (1, 74) = 3.529, p = .064).
Figure 2 shows type-token, lemma-token and lemma-type ratios
by age group. The type-token ratio is significantly affected by age (F
(4, 74) = 10.423, p = .000), it decreases from elementary school to
university but it increases in the teachers’ group. This result indicates
a higher lexical diversity in the oldest age group, so an interaction
both between genre and age (F (4, 74) = 5.349, p = .001) and between
modality and age (F (4, 74) = 4.608, p = .002) is found. Regarding the
lemma-token ratio, it trends similarly: there is a significant effect of
age (F (4, 74) = 7.310, p = .000), there is an interaction between genre
and age (F (4, 74) = 4.070, p = .005) and between modality and age (F
(4, 74) = 4.422, p = .003). Finally, the lemma-type ratio only shows an
almost significant effect of age (F (4, 74) = 2.252, p = .071).
174 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
Figure 1. Type-token-lemma ratios by text type.
Figure 2. Type-token-lemma ratios by age group.
Indicators of lexical growth for a Catalan L1 corpus 175
4.3. Lexical density
Table 4 presents lexical density by text type; roughly, almost half of
the tokens used on each text type are lexical tokens. Expository texts
have significantly higher lexical density than narrative ones, (F (1, 74)
= 30.151, p = .000). No significant effect of modality is found. By text
types, lexical density is higher in spoken expository texts than in
spoken narrative, but in written narrative is higher than in written
expository texts. There is a significant interaction between genre and
modality (F (1, 74) = 4.524, p = .037).
Table 5 presents the results by age group. Lexical density is
significantly affected by age (F (4, 74) = 10.193, p = .000). No
interactions between genre and age or between modality and age are
found. We can observe from the counts of lexical density that, except
for the secondary school group, lexical density shows a developmental
pattern. However, since the mean lexical density of each age group
increases gradually, we run Bonferroni post-hoc comparisons in order
to determine between which age groups significant differences are
found. Regarding discourse genres, significant differences in
expository texts are between the youngest and the oldest groups, that
is, between elementary school (M = .484, SD = .042) and teachers (M
= .527, SD = .032) (d = .043); and also between secondary school (M
= .470, SD = .036) and both university (M = .509, SD = .020) (d =
.039) and teachers (d = .057). For narrative texts, Bonferroni post-hoc
analyses show that significant differences are to be found between
elementary school (M = .431, SD = .044) and the three oldest groups,
that is, high school (M = .469, SD = .029) (d = .038), university (M =
.493, SD = .018) (d = .061) and teachers (M = .488, SD = .021) (d =
.056); and also between secondary school (M = .446, SD = .048) and
both university group (d = .046) and teachers (d = .041). As for
modality, significant differences in spoken texts are found between the
youngest and the oldest groups, that is, between elementary school (M
= .460, SD = .028) and both university group (M = .506, SD = .021) (d
= .045) and teachers (M = .505, SD = .025) (d = .044); and also
between secondary school (M = .456, SD = .038) and both university
group (d = .049) and teachers (d = .048). Finally, for written texts,
significant differences are found between elementary school (M =
176 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
.455, SD = .040) and the two oldest groups, that is, university group
(M = .496, SD = .021) (d = .041) and teachers (M = .509, SD = .025)
(d = .054); and also between secondary school (M = .459, SD = .035)
and teachers (d = .049).
Counts
M
SD
Spoken
Expository
Tokens
30,315
383.73
406.69
Lexical tokens
15,071
190.92
210.85
Lexical density
.497
.487
.040
Written
Expository
Tokens
16,929
214.29
143.39
Lexical tokens
6,960
107.06
73.19
Lexical density
.411
.495
.044
Spoken
Narrative
Tokens
22,151
280.39
275.56
Lexical tokens
10,527
133.53
138.20
Lexical density
.475
.463
.042
Written
Narrative
Tokens
14,686
185.90
104.56
Lexical tokens
6,897
86.25
52.78
Lexical density
.469
.455
.053
Table 4. Lexical density by text type.
Counts
M
SD
Elementary
School
Tokens
11,223
140.28
57.63
Lexical tokens
5,182
64.23
27.36
Lexical density
.461
.457
.031
Secondary
School
Tokens
12,200
160.52
71.70
Lexical tokens
4,122
73.96
34.42
Lexical density
.337
.458
.031
High School
Tokens
26,082
326.02
192.00
Lexical tokens
12,627
157.72
97.27
Lexical density
.484
.481
.022
University
Tokens
19,627
490.67
282.00
Lexical tokens
9,902
247.32
142.20
Lexical density
.504
.501
.016
Teachers
Tokens
14,949
373.72
75.37
Lexical tokens
7,590
190.82
41.95
Lexical density
.507
.507
.021
Table 5. Lexical density by age group.
Indicators of lexical growth for a Catalan L1 corpus 177
4.4. Word length
Figure 3 shows the results of word length by text type. Word length
shows significant differences between the genres and modalities
studied. Regarding the differences between genres, expository texts
have a higher word length than narrative texts, there is a significant
effect of genre (F (1, 74) = 19.614, p = .000). As for modality, written
texts have a higher word length than spoken texts, and significant
differences are found (F (1, 74) = 68.432, p = .000). By text type,
written expository texts have the longest words, followed by spoken
expository, written narrative and spoken narrative; there is a significant
interaction between genre and modality (F (1, 74) = 4.478, p = .038).
Figure 4 illustrates the word length results by age group. Word
length is significantly affected by age (F (4, 74) = 3.728, p = .000).
There is also a significant interaction between genre and age (F (4, 74)
= 4.707, p = .002) as well as between modality and age (F (4, 74) =
7.633, p = .000). This measure shows a clear developmental pattern,
because it increases throughout the subjects’ age, presenting a highest
increase between university and teachers groups.
Figure 3. Word length by text type.
178 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
Figure 4. Word length by age group.
4.5. Productivity of verbs
4.5.1. Diversity of verbal lemmas
Table 6 shows the diversity of verbal lemmas by text type. No
significant effect of genre is found. Written texts have higher diversity
of verbal lemmas than spoken ones (F (1, 74) = 114.586, p = .000). By
text type, written expository is the text type with the highest diversity
of verbal lemmas, .658, then spoken expository, with a diversity of
.516, closely followed by written narrative, .508, and finally, spoken
narrative is the text type with the lowest diversity of verbal lemmas,
.463. Table 7 shows the diversity of verbal lemmas by age group.
There is a significant effect of age, (F (4, 74) = 4.117, p = .005). The
results show that diversity increases gradually, except for the
university group. There is a significant interaction between genre and
age (F (4, 74) = 3.659, p = .009) as well as between modality and age
(F (4, 74) = 5.549, p = .001).
Indicators of lexical growth for a Catalan L1 corpus 179
Counts
M
SD
Spoken
Expository
Verbal lemmas
635
77.61
83.78
Different verbal lemmas
328
28.63
18.64
Diversity
.516
.471
.157
Written
Expository
Verbal lemmas
407
40.59
25.77
Different verbal lemmas
268
22.76
12.24
Diversity
.658
.597
.116
Spoken
Narrative
Verbal lemmas
621
61.09
53.90
Different verbal lemmas
288
24.27
14.65
Diversity
.463
.454
.125
Written
Narrative
Verbal lemmas
584
39.92
20.64
Different verbal lemmas
297
22.16
10.71
Diversity
.508
.575
.115
Table 6. Diversity of verbal lemmas by text type.
Counts
M
SD
Elementary
School
Verbal lemmas
323
33.02
14.57
Different verbal lemmas
154
16.76
5.87
Diversity
.476
.573
.096
Secondary
School
Verbal lemmas
301
35.14
16.81
Different verbal lemmas
151
17.23
6.41
Diversity
.501
.555
.089
High
School
Verbal lemmas
640
68.35
37.14
Different verbal lemmas
327
27.91
8.67
Diversity
.510
.529
.082
University
Verbal lemmas
597
96.12
54.21
Different verbal lemmas
285
37.92
14.13
Diversity
.477
.529
.035
Teachers
Verbal lemmas
546
67.30
11.05
Different verbal lemmas
293
33.17
6.10
Diversity
.536
.605
.033
Table 7. Diversity of verbal lemmas by age group.
4.5.2. Productive use of inflectional morphology of verbs
The productive use of inflectional morphology by text type presents
no significant differences between discourse genres and modalities of
production. Verbal lemmas with the least productive use of
inflectional morphology, that is, verbs which appear in only one tense,
180 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
represent more than a half of the total in all text types. Spoken
narrative is the text type which shows a more productive use of
inflectional morphology, verbs like anar ‘to go’ and dir ‘to say’ are
used in 11 different tenses. Written expository is the text type which
has less productive use of inflectional morphology, verbs like fer ‘to
do’ and veure ‘to see’, are used in seven different tenses.
The productive use of inflectional morphology by age group
presents no significant differences. More than a half of verbal lemmas
have the least productive use of inflectional morphology, they appear
in only one tense. Elementary school is the age group that shows a less
productive use of inflectional morphology, for example, the verb dir
‘to say’ is used in eight different tenses. Secondary school group has a
similar productive use of inflectional morphology, for instance, the
verb tenir ‘to have got’ is used in nine tenses. The oldest age groups
have more productive use of inflectional morphology, for example,
the verb dir ‘to say’ is used in 11 tenses in all these three groups.
Verbal lemmas which have a higher productive use seem to be
high frequency verbs (e.g. dir ‘to say’, fer ‘to do’, tenir ‘to have got’).
By contrast, verbal lemmas with a lower productive use seem to be
low frequency verbs (e.g. traumatitzar ‘to traumatize’, senyorejar ‘to
dominate’, violentar ‘to embarrass’). We have looked up in the DdF
(Diccionari de Freqüències, ‘Frequency List’) (Rafel 1998) of the
Institut d’Estudis Catalans, which is based on the CTILC (Corpus
Textual Informatitzat de la Llengua Catalana) and includes literary
and non-literary written texts in Catalan, in order to compare the most
frequent verbal lemmas of the GRERLI-CAT1 corpus with the
frequency of the same verbal lemmas in the CTILC corpus. Table 8
shows the 10 most frequent verbal lemmas (from highest to lowest) of
the GRERLI-CAT1 corpus on the first column; the second represents
their absolute frequency, that is, the number of occurrences of tokens
of this verbal lemma related to the total of tokens of the GRERLI-
CAT1 corpus (84,081). The third column shows their relative
frequency, that is, the percentage of the lemma representation to the
total of tokens of the corpus. In the fourth column, the same verbal
lemmas are presented from highest to lowest frequency, according to
the DdF; the fifth column represents their absolute frequency related
to the total of tokens of the CTILC corpus (52,375,044); and, finally,
Indicators of lexical growth for a Catalan L1 corpus 181
the sixth column shows their relative frequency. As we can see, the
results are quite similar, if we take into account the huge difference
between the number of tokens of both corpora. This confirms that
verbs with more productive use are the most frequently used.
Verbal lemma
GRERLI-CAT1
Absolute
frequency
Relative
frequency
Verbal lemma
CTILC
Absolute
frequency
Relative
frequency
dir ‘to say’
48
3.967
ser ‘to be’
1,000,352
1.951
fer ‘to do’
44
3.636
fer ‘to do’
338,070
.659
ser ‘to be’
37
3.058
tenir
‘to have got’
236,795
.462
anar ‘to go’
31
2.562
dir ‘to say’
222,380
.433
haver ‘to have’
26
2.149
poder ‘can’
211,688
.413
passar
‘to happen’
23
1.901
anar ‘to go’
114,340
.223
poder ‘can’
23
1.901
estar ‘to be’
111,112
.216
tenir
‘to have got’
17
1.405
passar
‘to happen’
64,499
.125
estar ‘to be’
14
1.157
començar
‘to start’
31,259
.060
començar
‘to start’
8
.661
haver ‘to have’
1,984
.003
Table 8. Comparative between the 10 most frequently used verbal lemmas of the
GRERLI-CAT1 corpus and the results of the DdF.
5. Conclusion
Our study enables us to distinguish the specific contribution of four
lexicon-related measures for the corpus-based study of lexical
development. The results show that word length offers the best
diagnosis of lexical development, genre and modality differentiation.
Lexical density is a good indicator of developmental changes and
genre differentiation, but not of modality differentiation. Lexical
diversity, by contrast, is a good indicator of developmental changes
and modality differences, but not of genre differentiation. Similarly,
182 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
diversity of verbal lemmas is also an indicator of developmental
changes and modality differences. Finally, productive use of
inflectional morphology is not appropriate for characterizing later,
school age language development of lexicon.
Word length appears as a valid measure to differentiate between
school level, discourse genres and modalities of production. There are
significant differences between age groups, effects of genre and
modality and also an interaction between genre and age, modality and
age and also between genre and modality. Word length shows a clear
developmental pattern from childhood to adulthood. Moreover, word
length is significantly higher in the expository genre than in the
narrative one, and in written compared to the spoken modality. In line
with other studies, such as Stömqvist et al. (2002), for Swedish
language, or Llauradó / Tolchinsky (in press), for Catalan language,
word length appears as the best diagnosis of development, genre and
modality differences.
Lexical density is a valid measure to differentiate between
school level and discourse genres. Significant differences between age
groups and an effect of genre are found, and there is also an
interaction between genre and modality. Lexical density yielded no
clear developmental pattern, like the ones found in previous studies
(Llauradó / Tolchinsky in press), though we found differences
between the youngest and the oldest groups. As for differences by
genre, expository texts are denser than narrative ones. However, and
unlike other studies in which lexical density showed significant
differences between written and spoken texts (Strömqvist et al. 2002),
we found no significant effect of modality of production.
Lexical diversity, measured by type-token, lemma-token and
lemma-type ratios, is a valid measure for revealing developmental
changes and modality differences. However, lexical diversity does not
account for genre differentiation. In both type-token and lemma-token
ratios, there is a significant effect of age and modality, and
interactions between genre and age, modality and age and also
between genre and modality. Regarding the lemma-type ratio, no
significant differences between genres, modalities and age groups are
found. The highest lexical diversity is found in the oldest age group, in
line with the findings of other studies (Berman / Verhoeven 2002;
Indicators of lexical growth for a Catalan L1 corpus 183
Strömqvist et al. 2002). Written texts have a higher type-token and
lemma-token ratios than spoken texts. This finding also corroborates
those found for type-token ratio for Swedish by Strömqvist et al.
(2002).
Finally, productivity of verbs functions halfway to our
purposes. As for diversity of verbal lemmas, there is a significant
effect of age and modality of production, and interactions between
genre and age and between modality and age are also observed.
Diversity of verbal lemmas increases gradually through age groups,
except for the university group. Written texts have higher diversity
than spoken ones. As for the productive use of inflectional
morphology of verbs, it does not seem as a valid measure to
differentiate between school level, discourse genres and modalities of
production. Nine-year-olds and adults display a similar use in the
production of inflectional morphology of verbs.
In sum, for our purposes, the most suitable measure is word
length, because it can characterise differences in all variables: age,
genre and modality. Lexical density is appropriate to differentiate
between ages and discourse genres, and lexical diversity reveals as
useful measure to differentiate between ages and modalities of
production. Finally, diversity of verbs can serve to differentiate
between ages and modalities, but productive use of inflectional
morphology of verbs is not a valid indicator of text construction
development in different genres and modalities.
The characteristics of the GRERLI-CAT1 corpus will allow us
to explore the type of lexicon used by different age groups, as well as
the syntactic organization of different types of texts, the textual
components of each discourse genre or the effect of the order of text
production (Cutillas in press), among others. Cross-linguistic
comparisons with the other languages of the main project will enable
us to analyse the similarities and differences between these languages,
as well. Moreover, comparisons with the same corpus for Catalan as
L2 (GRERLI-CAT2) will provide us with valid data to characterise
Catalan language development (L1) and language learning (L2).
184 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
Acknowledgements
Data for this paper were gathered within the I+D research project
Hacia el dominio experto de la lengua: estudio comparado del
desarrollo del repertorio lingüístico nativo y no nativo en castellano y
catalán ‘Becoming an expert user of language: a comparative study of
the development of native and non-native linguistic repertoire in
Spanish and Catalan’ funded by the Spanish Ministry of Science and
Innovation, reference code EDU2009-08862.
References
AnCora-CA corpus. <http://clic.ub.edu/corpus/ancora/>
Aparici, Melina 2010. El desarrollo de la conectividad discursiva en
diferentes géneros y modalidades de producción [The
Development of Discursive Connectivity in Distinct Genres and
Modalities of Production] (Unpublished PhD dissertation).
Barcelona: Universitat de Barcelona.
Berman, Ruth A. / Slobin, Dan 1994. Relating Events in Narrative: A
Crosslinguistic Developmental Study. Mahwah: Lawrence
Erlbaum Associates.
Berman, Ruth A. / Verhoeven, Ludo 2002. Cross-linguistic
Perspectives on the Development of Text Production Abilities:
Speech and Writting. Written Language and Literacy. 5/1, 1-44.
Boix, Emili / Alturo, Núria / Perea, Maria P. 2002. Corpus del Català
Contemporani de la Universitat de Barcelona (CUB). A General
Presentation. Romanistische Korpuslinguistik. Korpora und
gesprochene Sprache. Tübingen: Gunter Narr, 155-170.
<http://www.ub.edu/ cccub/>.
CesCa corpus. <http://clic.ub.edu/corpus/corpus/cesca/>
CICA corpus. <http://www.cica.cat/>
Indicators of lexical growth for a Catalan L1 corpus 185
Civit, Montserrat 2003. Criterios de desambiguación morfosintáctica
de corpus en español [Morphosyntactic Disambiguation
Criteria of Corpora in Spanish]. Alicante: Sociedad Española
para el Procesamiento del Lenguaje Natural.
CTILC corpus. <http://ctilc.iec.cat/>
Cutillas, Laia In press. Parlar per escriure o escriure per parlar? [Talk
for Writing or Write for Talking?]. In L. Tolchinsky (ed.) Cap a
una explotació didàctica dels corpus lingüístics [Towards an
Educational Explotiation of Linguistic Corpora]. Barcelona:
Horsori.
EAGLES 1996. Recommendations for the Morphosyntactic
Annotation of Corpora. <http://www.ilc.cnr.it/EAGLES96/
annotate/annotate.html>.
Granger, Sylviane / Wynne, Martin 2000. Optimising Measures of
Lexical Variation in EFL Learner Corpora. In Granger,
Sylviane / Wynne, Martin (eds) Corpora Galore: Analyses and
Techniques in Describing English. Amsterdam: Rodopi, 249-
257.
Llauradó, Anna / Tolchinsky, Liliana In press. The Growth of the
Text-embedded Lexicon in Catalan from Chilhood to
Adolescence. First Language.
Llauradó, Anna / Martí, Maria Antònia / Tolchinsky, Liliana 2012.
Corpus CesCa: Compiling a Corpus of Written Catalan
Produced by School Children. International Journal of Corpus
Linguistics. 17/3, 428-441.
MacWhinney, Brian 2012. The CHAT Transcription Format.
<http://childes.psy.cmu.edu/manuals/chat.pdf>.
Malvern, David / Richards, Brian / Chipere, Ngoni / Durán, Pilar
2004. Lexical Diversity and Language Development.
Quantification and Assessment. Hampshire: Palgrave
MacMillan.
Nesselhauf, Nadja 2004. Learner Corpora and their Potential for
Language Teaching. In J. Sinclair (ed.) How to Use Corpora in
Language Teaching. Amsterdam: John Benjamins, 125-152.
Rafel, Joaquim 1998. Diccionari de freqüències. [Frequency List].
Barcelona: Institut d'Estudis Catalans.
186 Laia Cutillas, Liliana Tolchinsky, Elisa Rosado, Joan Perera
Rafel, Joaquim / Solanellas, Josep M. 1986. El corpus textual
automatizat de la llengua catalana. Actas de las II Jornadas
Españolas de Documentación Automatizada. 147-161.
Read, John 2000. Assessing Vocabulary. Cambridge: Cambridge
University Press.
Serrat, Elisabet / Sanz-Torrent, Mònica / Bel, Aurora 2004.
Aprendizaje léxico y desarrollo de la gramática: Vocabulario
verbal, aceleración morfológica y complejidad sintáctica
[Lexical Learning and Grammar Development: Morphological
Productivity, Syntactic Complexity, and Verb Vocabulary
Learning]. Anuario de Psicología. 35/2, 221-234.
Strömqvist, Sven / Johansson, Victoria / Kriz, Sarah / Ragnarsdóttir,
Hrafnhildur / Aisenman, Ravid / Ravid, Dorit 2002. Toward a
Cross-linguistic Comparison of Lexical Quanta in Speech and
Writing. Writen Language and Literacy. 5/1, 45-67.
Taulé, Mariona / Martí, Maria A. / Recasens, Marta 2008. AnCora:
Multilevel Annotated Corpora for Catalan and Spanish.
Proceedings of 6th International Conference on Language
Resources and Evaluation, 96-101.
Tolchinsky, Liliana 2004. The Nature and Scope of Later Language
Development. In R. Berman (ed.) Language Development
across Childhood and Adolescence. Amsterdam: John
Benjamins, 233-248.
Torruella, Joan 2009. Los ejes principales en el diseño de un corpus
diacrónico: El caso del CICA [The Principal Axes in the Design
of a Diachronic Corpus: the case of CICA]. In Cantos, Pascual /
Sánchez, Aquilino (eds) A Survey on Corpus-based Research /
Panorama de Investigaciones Basadas en Corpus. Asociación
Española de Lingüística del Corpus, 21-36.
