A Grammatical Coding and Analysis System for Language Data from Normal and Brain-Damaged Children

A Grammatical Coding and Analysis
System for Language
Data from Normal and Brain-Damaged
Children
1. Background
This abstract will describe a grammatical coding
and analysis system devised to study language
development in normal and brain-damaged children, but which can be used to analyze mature as
well as developing grammars.
The research project for which this system was
developed investigates the lateralization and localization of language in its development. This research program studies language development in
children who have catastrophic; i.e., medically
intractable epilepsy for which they undergo surgical resection of the diseased tissue. The surgeries
range from unilobar resections (e.g., temporal lobectomy), to multilobar resections (e.g., temporalparietal-occipital lobectomy or, in more extreme
cases, hemidecortication (removal or disconnection of the entire cortex, often referred to as hemispherectomy)). The effects of disease and removal of different parts of the left and right
hemisphere at different ages are examined and
compared – left vs. right, one age vs. another, and,
importantly, brain-damaged child vs. normal developing child.
The research focuses on several questions, including: 1) the capacity of each hemisphere alone to
subserve lexical and grammatical development (a
comparison of left-hemispherectomied and righthemispherectomied children with each other and
with normally developing children), 2) the development of lateralization and localization of grammar (specifically, syntax and morphology) as opposed to lexicon, 3) the effects of localized brain
damage on specific functional subsystems of the
grammar; namely, the D(eterminer)-system, the
I(nflectional)-system, and the C(omplementizer)-
system, 4) the effects of localized brain damage on
lexicon – the establishment of a mental dictionary
of content words and their interrelations – as opposed to syntax and morphology, and 5) maturational constraints on grammar acquisition, again
syntax and morphology. The research is part of a
multi-disciplinary investigation involving the study of psychopathology, neurophysiology and neuropathology as well as linguistics, and other research questions are directed at better
understanding any relationship between linguistic
development and function in other areas. These
research questions include whether there is a direct
relationship between thought disorder and developmental linguistic deficits and whether we find
an association between specific patterns of linguistic delay or anomalies and specific neuropathology.
As can be seen, many of our research questions
can only be answered in relation to a comparison
with normal language development. However, the
patterns and range of what is attested in lexical
development and the acquisition of morphology
and syntax in normally developing children have
not yet been fully established. Our study thus
involves considerable data collection and analysis
of language from normally developing children to
enhance the data base which we can then use as an
appropriate metric against which the brain-damaged children in our study can be compared.
The study is a five year, longitudinal study, designed so that we can document language development; i.e., change over time, not just language
performance at a single point in time. For the
study, documentation and assessment of linguistic
performance for each “surgical” child is carried
out pre-operatively, at 6-months post-surgery, 12-
months post-surgery, and then at yearly intervals.
For each of these data points, data from a normal,
matched control are collected.
Both formal test performance and observational
data are collected. However, the major data source
for our evaluation of the language performance of
each child in the study is the language sample. It
is for the analysis of these spontaneous speech
samples that our coding and analysis procedures
were developed.
2. Coding System
Below is a brief description of our coding and
analysis conventions. These conventions are theoretically grounded in and motivated by Government and Binding theory as outlined by Chomsky,
1981, 1986; Pollack, 1989; and others. They arose
from the need to capture morphological and syntactic distinctions and generalizations than are afforded by the present CLAN commands of the
CHILDES database. Insofar as our coding system
uses the existing CHAT transcription format and
is amenable to existing CLAN commands, we
offer it as a useful extension to existing CHILDES
capabilities.
Our coding scheme adds to the speaker tier of the
CHILDES transcript system three additional tiers
for morphological, syntactic and lexical coding.
2.1 The Morphological Tier
On the morphological tier (%mor), morphemes
related to the functional heads C, I and D are coded
as such and labeled with codes identifying the
specific structures involved. See, for example, (1)
and (2) below.
(1) she is a Subject PROnoun, its nominative
case-marking received via and so related to the
functional head I[nfl], and is thereby coded:
IPROS|she.
(2) my, the POSsessive Determiner related to the
functional head D[et] is coded: DPOSD|my.
We also code for phonetically overt bound morphology on this tier. Stems and affixes are divided
by “-”, as in (3).
(3) Comes is coded: IF|come-s, where IF corresponds to a Finite form of the verb related to
functional head I.
So the sentence She comes would be coded as in (4).
(4) *CHI: she comes .
%mor: IPROS|she IF|come-s
Other free grammatical morphemes are also coded
by syntactic category on this tier (e.g., Preposition,
PROnoun, etc.), and utterance length is generated
on the %mor tier as well. Thus, all and only the
utterances and words and morphemes within them
to be included in such counts are put on this tier.
Errors related to all structures coded are identified
by “=”. Omissions, misselections, and overinsertions of morphemes are all captured through this
scheme as illustrated by (5) – (7):
(5) Omissions receive the code “=0x” where x is
obligatory but omitted.
(6) Misselections receive the code “=y” if y is
there instead of x.
(7) Overinsertions are coded as “-x-=x” if there
are two markings of x instead of one.
It is important to note that, although this tier is
labeled the “morphological” tier, its label should
not be taken literally. Rather, our morphological
tier should be understood as a place for coding
syntactic as well as morphological information as
specified above, designed to meet the principle
objectives of including but going beyond the coding of bound and free morphemes, to code a) the
functional categories C, I, D and their subcategories, and b) errors related to these functional categories and their subtypes (i.e., omissions, misselections and overinsertion).
2.2 The Syntactic Tier
The syntactic tier (%syn) is designed as the place
for coding constituent structure (including types
of embedding and internal phrasal and clausal
structure), constituent order (capturing linear order and movement), and related errors. Each syntactic phrase is labeled with category and grammatical function labels as illustrated by (8) - (10):
(8) SNP = Subject Noun Phrase (category = NP,
grammatical function = Subject)
(9) MOD = Modal
(10) CADJ = ADJunct clause, complementizer
in C (projects up to C)
Embedded clauses are enclosed in parentheses ( )
and codes for embedding types in square brackets
( [ ] ). Verbal morphology is also included on this
tier, using the same codes as used on the morphological tier. An example coding of a well-formed
utterance is shown in (11).
(11) *CHI: She can’t sing if he’s in the room.
%syn: SNP MOD NEG V ( [CADJ] SNP
IS PP )
Errors (e.g., order errors, omission errors) are also
coded, using “=” to signal errors as on the morphological tier. (12) illustrates the coding of an
omission error.
(12) *CHI: *He going
%syn: SNP IS|=0 V-ing
Our system for coding syntactic structure on this
tier allows for analysis of the coocurrence of grammatical structures (e.g., null subjects and finite
verb marking).
2.3 The Lexical Tier
The lexical tier (%lex) should be understood as the
tier for coding types, tokens, and errors of the
major lexical category words (N’s, V’s, ADV’s,
and ADJ’s) in each utterance of a speech sample.
All of the morphemes which are part of the same
lexical entry are placed, without spaces between
them, in the same listing. An “=” is again used to
signal an error. The lexical coding is illustrated in
(13) below.
59
(13) *CHI: My father poured dinner for us.
%lex: N|father V|=pour N|dinner
This tier allows us to examine the structure, size
and productivity of the speaker’s lexicon. Omitted
words would be captured on one of the other tiers
defining the resultant grammatical error.
This coding system is still undergoing modification. We are forced to add to or modify our current
system to accomodate data which deviate in unexpected ways from the normal, adult grammar. By
designing our system to describe and analyze disordered as well as normal language, however, we
have developed a system which is useful for linguistic analysis of all kinds of speakers, including
normal children, children with language disorders,
adults with acquired aphasia and, of course, normal mature speakers.
3. Analysis System
Freq.exe, one of the CLAN programs associated
with the CHILDES Project1
, provided the basis for
an elaborated computer analysis system which we
designed to answer specific research questions
regarding various subsystems of the grammar.
While freq.exe is a powerful tool for counting
frequencies of words, morphemes or codes, research questions posed by our project required that
a variety of very different morphological, syntactic, and lexical codes be counted as evidence regarding a single subsystem of the grammar. With
respect to the I-system, for instance, our coding
system marks a subject pronoun she as IPROS|she,
and errors in IPROS as IPROS|=she. Counts of
IPROS in the language samples could be obtained
using the standard freq.exe command shown in
(14); however, the error counts required two
freq.exe commands, as shown in (15).
(14) freq +t%mor -t* +s “*IPROS|*” acoded.ext
>output1.ext
(15a) freq +t%mor -t* +s “IPROS|=*” acoded.ext
>output2.ext
(15b)freq +t%mor -t* +s “=IPROS|*” acoded.ext
>output3.ext
(15a) is an error with respect to the functional
category itself, while (15b) is an error with respect
to the element within that functional head.
For the I-system alone, occurrences of fifteen different codes and their errors needed to be counted
individually and sorted for inspection. Because we
code for a variety of errors in each category, two
or three freq.exe commands would be necessary
for each error subtotal, depending upon the range
of errors coded in the data for each category.
Occurences of error obtained in freq.exe searches,
such as those in (15), would have to be added
together in order to account for the total number
of errors in each category under analysis. In all,
forty-seven freq.exe searches would be required
for each data point. Besides amounting to an arduous and time-consuming task, the conventional
freq.exe method left us with the need to subtotal
error counts for each category by hand, thus introducing an opportunity for miscalculations in our
results.
Therefore, in order to simplify and streamline
analyses for the I-system, we created a program
called i-sys.exe which performs four basic operations, as outlined in (16).
(16) Each time i-sys.exe is run, it
(a) calls freq.exe forty-seven times to conduct relevant I-system searches on
coded files, creating a separate freq.exe
file containing search results for each
iteration;
(b) reads the forty-seven files created in (a)
and extracts the “total number of
words” (codes, actually) in each; and
(c) computes results obtained in (b) to
compile a final report called i-report.ext.
(d) calculates the number of morphemes,
number of utterances, and mean length
of utterance (MLU)
The forty-seven freq.exe output files are all compiled into a single file called freq-out.ext and
deleted from the disk. The analysis procedure is
completed in a run time of approximately 13 seconds under DOS, on a 80 MHz. Pentium computer.
The collection of reports generated allows easy
inspection of various aspects of the I-system, including the proportion of errors in each. Comparable “reports” are generated for the D- and C-systems, as well as closed class items.
4. Critical Assessment and Conclusions
This methodology and these analysis tools free us
from the alienating labor of manually counting
tokens in our data sets and provide us with a rich
and accurate basis for interpreting our data. However, although this system is appropriate for our
research objectives and of great assistance to us in
our work, as theoretical insights and the data demand, we will continue to modify and improve it.
Notes
1
See Brian MacWhinney, The CHILDES Project:
Tools for Analyzing Talk (New Jersey: Lawrence
Erlbaum, 1991).