harris.doc - PsyLing

Transfer and Competition in
Second Language Learning Brian MacWhinney
Carnegie Mellon University
Recent work has explored the application of the Competition Model
(MacWhinney & Bates, 1989) to the study of second language acquisition.
In making this extension, it is important to distinguish between transfer
from L1 and direct learning of L2. Both processes can be analyzed in
terms of the constructs of cue reliability, cue cost, and form-function
mappings. The model predicts certain typical varieties of transfer
during the process of phonological, syntactic, and lexical learning. In
the attempt to maximize the transfer of L1 structures the learner uses a
variety of complex learning strategies. In areas where transfer is
poorly supported, the learner acquires L2 structures directly. Cue
reliability and cue cost estimates can also be used to characterize the
direct acquisition of L2 structures.
Psycholinguists come in two types. There are those who study
adults and those who study children. The psycholinguists who study
children care about "acquisition" and the psycholinguists who study
adults care about "processing." Until recently, these two groups have
acted much as separate "modules." This separation is most unfortunate,
since we know that acquisition influences processing and that processing
influences acquisition. Fortunately, the study of second language
acquistion provides a way to dissolve the barriers between these
modules. Adult second language learners and adult bilinguals make
excellent experimental subjects. Unlike children (Friederici, 1983;
1983; Tyler & Marslen-Wilson, 1981), adults can read printed words on a
computer screen and make quick and well-controlled judgments in reaction
time experiments. Because it is relatively easy to acquire data on
language processing in adult second language learners and bilinguals, we
can use this population to broaden our understanding of the changes in
language processing that occur during language acquisition (Kilborn,
1989). The study of adult second language learning also allows us to
correct another limitation in the scope of adult psycholinguistics.
Virtually all of the current edifice of adult psycholinguistics is built
upon data derived from the study of language processing in English-
speaking college freshmen. This anglocentric bias is bound to lead to a
misleading view of the language-making capacity. Child language
researchers have already begun to escape from this bias. The ground-
breaking crosslinguistic developmental research of Slobin (1985) and his
colleagues has vastly expanded our understanding of what it means to
acquire a human language. Adult psycholinguistics has only recently begun to escape from its
anglocentric straight-jacket. Within the context of a model of sentence
processing called the Competition Model, MacWhinney and Bates (1989)
have opened the doors of psycholinguistics to a wider array of
crosslinguistic data, including adult sentence processing, language in
aphasia, and second language learning. Although the Competition Model
(Bates & MacWhinney, 1982) was not originally based on data from second
language acquisition, its crosslinguistic developmental orientation
seems to make it well-suited for use in this area too. At the same
time, it is clear that data on the learning of foreign languages can
play an important role in the testing and elaboration of the Competition
Model. Form-Function Mappings The fundamental idea underlying the Competition Model is simple
and rather traditional. The model takes as its starting point the
Saussurean vision of the linguistic sign as a set of mappings between
forms and functions. Forms are the external phonological and word
order patterns that are used in words and syntactic constructions.
Functions are the communicative intentions or meanings that underly
language usage. In the Competition Model, each lexical item or
syntactic construction can be understood as a form-to-function mapping.
Take the word "bat" as an example. The functions for this word involve
the expression of the various semantic properties of the animal, along
with its visual and auditory images. The form of the word is the set of
phonological cues contained in the sound sequence /bAt/. On the
syntactic level, a similar relation holds. Structures such as preverbal
positioning or verb agreement marking are treated as forms. These forms
are mapped to functions such as agency, topicality, perspective, first
mover, causer, volitional agent, and so on. In addition to the correlations between particular subsets of
forms and subsets of functions, there are also correlations within the
overall set of functions and within the overall set of forms. We can
call these function-function correlations and form-form correlations.
On the functional level, it is generally true that topics can also be
animate, definite, given, and so on. These correlations are
reflections of certain real correlations between properties of the world
in which we live. Because the functions we choose to talk about are so
highly correlated in real life, the the forms we use to talk about these
functions also become highly correlated. This makes it so that no
single form expresses any single function and the relation between forms
and functions is both fluid and robust. There are also important
correlations on the level of forms. For example, words that take the
article "the" also are capable of taking the plural suffix, and so on. We speak of these various mappings as correlations because we know
that all categories are imperfect and subject to category leakage
(MacWhinney, 1989). For example, the correlation between preverbal
positioning and agency in English breaks down in the passive and the
imperative. Similarly, the correlation between plural marking and
semantic plurality breaks down with words like "pants" or "faculty." Cues and Cue-Validity Our analysis so far has kept close to the basic Saussurean concept
of a form-function mapping. To recast this thinking in terms of a
processing model, we will have to adapt some of the terminology.
Instead of forms, we can talk about the "cues" used by the listener to
facilitate the activation of alternative functions or "competing
construals." For example, the individual phonological segments in the
word "bat" can each be viewed as cues to activation of the meaning
underlying "bat." The cues in the first two segments would also
activate words like "bad" and "bag" and so on. For a fuller discussion
of the ways in which cue match can facilitate activation see MacWhinney
and Anderson (1986). Note, however, that the equation of "cue" with
"form" holds only for comprehension. When we switch to thinking about
sentence production, we need to think of the underlying functions as
cues, and the actual forms being selected as "competing forms." This
use of the term "cues" allows us to draw parallels between processing
and acquisition in comprehension and production. In both processes, the
activation of certain cues as inputs is what leads to the final
selection between competing outputs. Most of the work on the Competition Model has focused on sentence
comprehension. The interaction of cues such as preverbal positioning,
animacy, case-marking, and subject-verb agreement has been modelled
mathematically in the Competition Model using maximum likelihood
techniques (McDonald & MacWhinney, 1989). The data modelled in these
studies come from experiments with real subjects in many languages and
at many age levels using sentences in which the various cues are placed
into "competition" with each other in an orthogonalized ANOVA design.
The maximum likelihood techniques make it possible to estimate the cue
strengths of particular cues. For example, in our studies of sentence
processing in English, Italian, German, French, and Hungarian, we have
been able to estimate the relative strengths of preverbal positioning,
subject-verb agreement, and animacy as cues to the function of "agency."
We have found that, in English, the preverbal positioning cue is
extremely strong and that the agreement and animacy cues are only of any
importance at all when there is no preverbal noun, as in VNN sentences.
In Italian, on the other hand, the agreement cue is far stronger than
the word order cue. Although both English and Italian are described as
SVO languages, the actual strengths of the basic cues to sentence
interpretation in these two languages are radically different. Perhaps the most important empirical claim of the Competition
Model is that cue strength in the adult native speaker is directly
proportional to cue validity. What is crucial about this claim is that
our cue validity measures are taken from actual text counts based on
spoken or written discourse, whereas our cue strength measures are
derived from experiments. This way of understanding the relation
between the learner and the environment avoids the circularities often
found in mathematical modelling in psychology. The idea is that, during
language learning, the child comes to appreciate the relative order of
cues in his language and to tune his cognitive system so that it
correctly mirrors the environment. At first, the child picks up cues
on the basis of their overall availability. At this ea