（ＪＩS shift software required to read Japanese title, publication information and abstract)
ダニエル・ストラック 「応用言語学における言語習得理論：神経学的視点から」 『早稲田大学語学教育研究所紀要』 （編集・発行：早稲田大学語学教育研究所、2002年3月、57号、1−31頁）
梗概"Theories of Learning in Applied Linguistics: A Neurobiological Perspective"
(c) 2002 Daniel C. Strack, Visiting Associate Professor, Institute of Language Teaching, Waseda University, Tokyo, Japan, April, 2001-March, 2002
(First Published in Bulletin of the Institute of Language Teaching, Waseda University, Vol. 57, March, 2002, pp.1-31. The English abstract has been appended to the first page of the article in this HTML format.)-----pg.1
In that presuppositions are often in place before research
concerning language learning is even addressed, when empirical evidence does not match
well with linguistic theory, it is difficult to resolve the disparities. A neurobiological understanding of language
processing can offer perspectives that may help to reconcile some apparent contradictions,
both for language learning, and with respect to other related issues in which such
conflict occurs. This paper is an attempt to
address a “neurobiological blind spot” in applied linguistics, and also to recommend deeper neurobiological
understanding to applied linguists that cannot harmonize their findings with mainstream
linguistic theory.
Historically,
crises caused by fundamental disagreements in academia have been precursors to
groundbreaking new approaches. According to
some scholars, applied linguistics is presently facing just such an identity crisis. Peter Skehan points to the “uneasy relationship” of theory and pedagogy,
noting that “pedagogical applications derived from theoretical
approaches have only a perfunctory quality, rather than being properly rooted in theory.”
(Skehan 1998: 2) With the
growing tension between the real world findings of corpus linguistics and the analysis
prevalent in mainstream linguistic theory, the situation looks to become even more
complex. (Widdowson 2000) How can applied
linguists hope to resolve conflicts when empirical research can so easily be subsumed by
theoretical debate? Even on the pedagogically
crucial question of whether conscious learning is possible, nothing can be taken for
granted. Richard Schmidt observes, “A hundred years of research in psychology and centuries of argumentation in
philosophy have not resolved the issue, and I cannot resolve it here.” (1995: 28) If such controversies
are left to psychology and philosophy, they may well prove intractable.
Neurobiological knowledge could be used to augment the findings
-----pg.1/pg.2
of
applied linguistics and mediate conflicts when evidence from psycholinguistic research
does not correlate with linguistic theory. There
is a “cognitive” tradition in
applied linguistics, but this tradition has drawn more from psycholinguistics than
neurobiology. Although neurobiological issues
have garnered sporadic attention, (e.g. Jacobs and Schumann 1992) cognitive scientist and
linguist Sydney Lamb notes, “For most linguists the
orientation of neurocognitive linguistics is still new and unfamiliar, even while the term
'cognitive' is being used with ever greater frequency.” (1999:
13)
As
linguists have depended principally upon psycholinguistics to supply “cognitive” insight, they might perhaps be excused
for their present unfamiliarity with neurobiology. Neuroscientist
and Nobel Prize winner Gerald Edelman is not as quick to defend the indifference he
perceives in psychology itself. Edelman
states, “psychology can no longer declare its autonomy from
biology, and it must always yield to biology’s findings.”
(1992: 177) On the other
hand, Edelman has praised the work of cognitive linguists, Ronald Langacker and George
Lakoff, and philosopher, Mark Johnson, stating that his own work on neurobiology “nicely complement[s] Langacker’s, Lakoff’s and Johnson’s work, providing essential
biological underpinnings for many of their proposals.” (1992:
252) This is not surprising in that the
values Lakoff and Johnson list as fundamental to cognitive linguistics recognize that “mind” cannot exist without “brain.” They
state that cognitive linguistics “seeks to use the discoveries
of second-generation cognitive science to explain as much of language as possible,”
(1999: 496) while providing an account that is “neurally realistic,” “based on converging
evidence from as many sources as possible,” and encompassing “empirical generalizations over the widest possible range of phenomena.”
(1999: 79-80)
-----pg.2/pg.3
This paper will focus on the biological processes that underpin language
learning, and show how neurobiological insight may help resolve some of the apparent
contradictions between standard linguistic theory and what is actually observed. Some problematic empirical evidence relating to
error and language learning will be reviewed in section A, followed by a brief summary of
the brain’s neurobiological underpinnings in section B. After showing how an understanding of neurobiology
can illuminate previously inscrutable language processes in section C, section D will make
a case for applied linguistics to proceed with a “cognitive”
approach that incorporates evidence not just from psycholinguistics,
but from neurobiology, as well.
It has been claimed that children learning their native languages are not usually exposed to negative input (active error correction). (Wexler and Culiver 1980) To account for the speed at which children learn their native languages, some linguists posit a Language Acquisition Device (LAD), an innate knowledge of linguistic "rules" which children use to make sense of language in the absence of negative input. If, indeed, a LAD eliminates the need for negative input then explicit error correction should have no place in the second-language classroom. In response to this claim, some researchers posit that child language acquisition takes place during a critical period of development and so comparison with older second language learners is not justifiable. Even so, if there is a LAD, even one not functioning at full capacity due to the fact that a critical period has ended, then the principles, conditions and rules of language might prove inaccessible to
-----pg.3/pg.4
conscious
awareness, rendering explicit error correction either irrelevant or inferior to implicit
learning. Already the issue of language
learning has become complex; we find three questions that have the potential to frame all
dialogue:
1.
Is there a critical period in language learning?
2.
If there is a critical period, how does it work?
3.
Is there biological evidence of a Language
Acquisition Device that might render explicit error correction irrelevant?
It is
important to identify how errors occur. If
different aspects of language are processed in fundamentally different ways, then any
theory of error would need to take these differences into account. The experimental psychologist Steven Pinker has
posited that “modules” in the
brain (namely phonology, lexicon, morphology, syntax and semantics) process the different
aspects of language. (Pinker 2000: 23) Some
applied linguists are explicit in their acceptance of modularity and base their
examinations entirely upon the theory; others, while avoiding explicit commitments, simply
phrase their arguments in modularity-friendly terms.
Two more questions emerge:
4.
Is the brain modular in its processing of language?
5.
Are the terms associated with modularity biologically meaningful?
While these two questions seem only to be two different ways of phrasing the same query, the distinct wordings are useful in that they allow the topic to be analyzed from two divergent perspectives. Question 4 frames the question in straightforward biological terms
-----pg.4/pg.5
while
question 5 addresses more subtle issues relevant to linguistic analysis. They are not entirely the same question and will
not necessarily entail identical answers.
Positing
a phonology module further complicates an already challenging issue. Ian Watson remarks: “The
subject of phonology is in itself so complex and riddled with controversy that it is
difficult to know where to start to find a conceptual framework that will permit bilingual
acquisition to be investigated.” (1991: 25) One aspect he mentions is the relationship of
speech perception and production: are they two sides of the same coin or two different
coins? In his paper on phonological
acquisition in bilingual children, he mentions a case in which empirical evidence strays
from theory:
All of the perceptual studies reported above had subjects who in their productions were, at least for the variables tested, impressionistically indistinguishable from monolinguals. Their productions were not, however, identical when measured instrumentally. (Watson 1991: 45)
Watson noted that these findings, if unmitigated, would imply that bilinguals’ systems are separated in respect to production but utilize a single common system for perception. Watson found this interpretation “unlikely” because it would imply that perception and production are separate; he explains that there must have been some problem with the experiment. Why would the dissociations implied by the evidence present a problem to be mitigated? Advocates of generative phonology recommend a single phonological module that facilitates both perception and production. (Pinker 2000: 110) In this case, the theoretical
-----pg.5/pg.6
assumptions
of modularity and generative phonology disallow a straightforward interpretation of the
empirical evidence. Momentarily setting
theoretical questions aside, one might ask:
6.
Is there any neurobiological evidence for either the unity or
separation of perception and production?
Clearly,
questions such as these will have great repercussions for a second language learning
environment. If the perception and production
aspects of language are unified, then student errors in one domain must be caused by some
kind of unexplained interference within the modular associations themselves, beyond the
student’s conscious control.
Consequently, if extensive language input does not solve the problem naturally,
nothing will. In contrast, if the two aspects
are separate, then problems with perception or production may potentially be addressed
locally and more or less irrespective of the counterpart ability. Furthermore, if either perception or production
should be distinct from the hidden unconscious phonological module, then conscious
learning becomes possible, in turn paving the way for noticing, and possibly explicit
error correction.
Temporarily
setting problems of modular interference aside and proceeding further into Watson’s examination, two new assertions are made concerning the perception and
production of language in bilingual infants:
As the child’s vocabulary increases, it rapidly becomes uneconomic to store words as wholes. At this point, words are analyzed into smaller units, probably first into syllables, then segments. (Watson 1991: 32)
-----pg.6/pg.7
Watson
gives no evidence for the first contention but these two points are often used in
conjunction to assert biological motivation for generative phonology. (e.g. Pinker 2000:
93-94) The second assertion includes
documentation, but the two claims are only noteworthy as they stand together. Without the implied biological motivation, the
second assertion could simply be taken as evidence for metalinguistic understanding.
Having
stated these two positions, Watson encounters another paradox. It seems that children can produce sounds that
have not yet been phonologically categorized. (1991: 30-32)
How is this a paradox? Normally,
perception is thought to drive production and so how can an effect outrun its supposedly
antecedent cause? Watson shrugs this apparent
contradiction off by saying that it doesn’t matter that these
two aspects appear to be separate in infants because the performance “does not translate usefully into phonological knowledge.” (1991: 30-31) The beginning of “linguistic activity” is thus deferred to a later
age, making any conflict with the principles of generative phonology irrelevant. A very practical question presents itself:
7. Does neurobiological evidence show any “economic” necessity to break words down into
syllables or segments?
In contrast to these assertions of an “economic” necessity for all language learners to break words down, Bialystok mentions anecdotal evidence from the second language classroom that apparently contradicts generative grammar theory: “Second language learners frequently
-----pg.7/pg.8
manage
to reduce the demands for analysis by the use of highly practiced and conventionalized “chunks” or patterns of language.” (1991: 127) This is problematic because a generative view asserts that
language is a discrete combinatorial system, which is “infinite,”
and “compositional.” (Pinker 2000: 365). True
linguistic learning cannot take place in the absence of unconscious analysis for
combinatorial use.
Bialystok
solves this problem by asserting that second language students only “appear” to be learning by using chunks. (1991:
127) Although the students may seem to be putting such unanalyzed “chunks” into practical use, these “meaning units” are not legitimate aspects of
language because they have not yet been analyzed apart for potential recombination. The analysis will occur internally and after the
fact (and so the rote memorization of these unwieldy “chunks”
does not serve a true linguistic purpose). Because it seems unlikely that students would
stubbornly display such effort to no practical end, another question comes to mind:
8.
Wouldn’t it be possible
for second language learners to use conventionalized “chunks”
for real communication even lacking combinatorial analysis?
While question 8 might at first glance appear not to be an issue of contention, it does, in fact, have crucial theoretical implications. Although many linguists would accept at least some use of unanalyzed “chunks” of language, opinions concerning the extent of their employment may be seen as a continuum between “limited utility” and “comprehensive use,” with many Universal Grammar advocates at the former end of the continuum and radical construction grammar advocates
-----pg.8/pg.9
at
the latter. Using neurobiology as a standard
to gauge the relative feasibility of “chunks” as opposed to “combinatorial analysis” is an exercise of crucial theoretical import.
For
the same reasons that “chunking” presents
a challenge to a combinatorial view of language, regression errors in language learning
also present problems for those who would advocate an unconscious LAD. Bialystok comments:
It is commonly observed that second-language learners demonstrate considerable variability in their apparent control over the forms of the new language. Correct forms seem to slip in and out of the learner’s speech, defying any accurate measure of progress with the second language. (Bialystok 1991: 136)
Children, as well, are observed to backtrack into mistakes, which poses a dilemma for a generative view of language. Once analysis has taken place, why would learners at any level make regression errors? Bialystok mentions a possible explanation: the LAD must have multiple levels at which to function, with symbolic representations, formal representations and semantic representations functioning independently, so that it can deal with an individual’s progress uniquely as the learner’s skills develop: “In time, each of these will become analyzed to a higher level.” (Bialystok 1991: 118) As some levels lag behind others in sophistication, errors occur. Pinker has a different explanation: “Human memory profits from repetition. If children have heard sang less often than adults have, their memory trace for it will be weaker and their ability to retrieve it will be less reliable.” (Pinker 2000: 197) This statement, however, implies that the LAD must be given the same
-----pg.9/pg.10
information
over and over before it will do its job, leaving the theoretically powerful universal
grammar at the mercy of frail human memory.
When
confronted with evidence counter to claims for an autonomous generative grammar, neither
scholar considered questioning the existence of a LAD; instead, the device itself needed
only to be scaled up for Bialystok or scaled down for Pinker. Momentarily setting aside these theoretical
explanations, the issue may be addressed without reference to any LAD:
9.
Is there any neurobiological explanation for the way in which both
young native speakers and second language learners regress into errors?
Looking
back over the questions suggested, it is apparent that all of the problematic evidence
conflicts with theoretical assumptions that were held from the outset. Furthermore, the conflicts are all mitigated
either by adjusting aspects of the theory or carefully redefining language itself; the
legitimacy of the starting hypotheses are never questioned. Not coincidentally, the
presuppositions being defended are not separate, unrelated issues, but rather
interconnected, supporting arguments in a theoretical pyramid.
We
will return to these questions again in section C and attempt to answer them from a
neurobiological perspective. Before that,
however, an overview of neurobiology with respect to development, memory and language will
introduce the main concepts and terminology necessary to answer these questions.
-----pg.10/pg.11
Evidence
will be taken from the work of neuroscientists and cognitive scientists that either study
neurobiology directly or emphasize neurobiological evidence in their research. These researchers include Mark F. Bear (Professor
of Neuroscience, Brown University), Barry W. Connors (Professor of Neuroscience, Brown
University), Antonio Damasio (Department of Neurology Head, University of Iowa’s College of Medicine), Terrence W. Deacon (Associate Professor of Biological
Anthropology, McLean Hospital, Harvard University), Gerald M. Edelman (Director of
Neurosciences Institute, Department of Neurobiology Chair, Scripps Research Institute),
Sidney Lamb (Professor of Linguistics and Cognitive Science, Rice University), Michael A.
Paradiso (Professor of Neuroscience, Brown University) and Giulio Tononi (Senior Fellow in
Theoretical and Experimental Neurobiology, Neurosciences Institute). While some of these scholars might disagree with
one another on more subtle matters such as mechanisms of consciousness or non-human
cognition, there is agreement on neurobiological fundamentals, including much of the
information presented in the following overview.
Specialized cells called neurons form the network that accomplishes the various goals of the body by transmission of electrochemical stimulus. A prototypical neuron has three basic parts: the soma (cell body), the axon (an output fiber) and dendrites (input fibers). Neurons are connected to other neurons, forming electrochemical circuits that consist of “conducting wires (the neurons’ axon fibers) and connectors, known as synapses (which usually consist of an axon making contact with the dendrites of another neuron).” (Damasio 1999: 324) Neurons generate nerve impulses that are called action potentials. While the
-----pg.11/pg.12
action
potentials of one neuron will be different from those of other neurons, the multiple
action potentials resulting from the firing of a single neuron are consistent in size and
duration. (Bear et al 2001: 74) Generally
evidencing greater similarity to a burning fuse than to an electrical charge, (Bear et al
2001: 89-91) these electrochemical impulses do not carry “information.” (Edelman 1992: 27) The spikes (action potentials) generated by
neurons are similar to the clicks of a Geiger counter in the presence of radiation; in
response to strong stimulus a neuron fires rapidly and weak stimulus will produce less
frequent or intermittent spikes. These
firings are not in any way “encoded” and are functionally distinct from the highly calibrated succession of
electrical pulses that travel along telegraph wires or through digital processing
equipment.
The network of neural subsystems instantiates itself according to the basic cellular processes of division, migration, death, adhesion and induction. Although the timing of these events is coordinated according to genetic constraints, “…individual cells, moving and dying in unpredictable ways, are the real driving forces” of neural development. (Edelman 1992: 60) The overall configuration of the brain is genetically coordinated but from early embryonic stages, “neurons extend myriads of branching processes in many directions” and connectivity is established at the synapse level as a result of individual development. (Edelman and Tononi 2000: 83) Neurons do not simply branch out to complete the system; a mature and functional neural system requires some little-used connections to be eliminated while more active connections are strengthened. Edelman has called this selection process Neural Darwinism. Deacon notes, “Nature prefers to overproduce and trim to match, rather than carefully monitor and coordinate the development
-----pg.12/pg.13
of
innumerable cell populations.” (1997: 197)
A
neuron will only generate an action potential at the near-synchronous firing of many
inputs coming from other neurons. (Deacon 1997: 202)
Furthermore, the magnitude of the input stimulus can increase the firing frequency of the action potentials; the more
intense the stimulus, the greater the chances that a connection will be selected to “live” and become entrenched. There are two rules of thumb that sum up this
neuronal selection process: “neurons that fire together, wire
together,” and “neurons that fire
out of sync, lose their link.” (Bear et al 2001: 731)
Linked together, neurons form somewhat localized
brain units, but “there are no single “centers” for vision, or language, or for that
matter, reason or social behavior. There are “systems” made up of several interconnected brain
units.” (Damasio 1994: 15) While these distributed systems
facilitate certain recognizable cognitive functions, the contribution of a given brain
unit to the operation of the system hinges not only on the structure of the unit but also
on its place in the system. (Damasio 1994: 15) Sharp functional distinctions between regions in
cortical processing do not exist. (Bear et al 2001: 648)
In neurological terms, when an object is perceived visually, there is a neural pattern (or mental image) that registers at various processing stages between the eye and the brain. Mental images need not be visual: auditory images, olfactory images, gustatory images and somatosensory images all leave their marks on the system in one form or another. (Damasio 1999: 318-319) As mentioned before, these images are not “contained” in a single action potential. They are formed through the correlation of the action potentials of multiple neurons into “maps.” Such mapping processes are crucial to the operation of
-----pg.13/pg.14
complex
brains. “Maps
relate points on the two dimensional receptor sheets of the body (such as the skin or the
retina of the eye) to corresponding points on the sheets making up the brain.” (Edelman 1992: 19) Damasio notes
that these mappings do not need to be point-for-point, utterly faithful facsimiles of the
original perception; the brain constructs maps according to its own parameters, (1999:
322) the only real constraint being that these parameters must be adaptive enough to
succeed within their specific system. (Edelman 1992: 204, 220).
These
mappings are not found in a single location of the brain but are “distributed” over a number of locations. (Damasio
1994: 106-107) Damasio describes how
disparate aspects of a conceptualization of your Aunt Maggie might be distributed
throughout the brain:
There are dispositional representations for Aunt Maggie’s voice in auditory association cortices, which can fire back to early auditory cortices and generate momentarily the approximate representation of Aunt Maggie’s voice […] Aunt Maggie as a complete person does not exist in one single site of your brain. She is distributed all over it, in the form of many dispositional representations, for this and that. (Damasio 1994: 102-103)
One great advantage of the human brain’s distributed memory system is its relative immunity to catastrophic loss if some neurons die. (Bear et al 2001: 749) It is true that drinking alcohol kills neurons, but one drink will not likely cause the drinker to forget the word “aardvark.” The highly parallel and redundant nature of mental images assures that it is impossible for one neuron to contain the word or for
-----pg.14/pg.15
a
single neuron to allow access to it. (Lamb 1999: 173)
Heavy drinking over a long period of time is another matter, as evidenced by the
widespread brain damage often caused by Korsakoff’s syndrome,
a neurological disease associated with chronic alcoholism. (Bear et al 2001: 760) One
disadvantage of the highly distributed memory system is relative instability of logic when
compared to the precisely specified logical determinacy of systems such as Turing
Machines. (Edelman 1992: 225) Indeed, the reason that computers commonly exceed human
capacities for chess and mathematical calculation is found in fundamental structural
difference rather than memory capacity; the machines were built to be rigorously logical
and humans were not.
Neural
control of bodily movement is facilitated by motor
sequences, which are also mappings. These
motor sequences are necessary for kicking a ball, playing the piano or speaking. Motor sequences that are related to these tasks
are “constructed or linked during consciously guided learning
until a smooth, apparently effortless sensorimotor loop is executed speedily, reliably and
unconsciously.” (Edelman and Tononi 2000: 188) While motor sequences, including speech, are
constructed using overt, conscious control, they are entrenched through repetition,
eventually becoming global mappings for largely unconscious coordinated action.
In the case of sound perception, there is a certain amount of preprocessing that happens even before the sound image reaches Wernicke’s area, one of the regions often associated with language comprehension. Intensity and frequency adjustments occur with the mechanical interaction of the tympanic membrane and the ossicles even before neural processing is initialized in the cochlea. Beginning with the cochlea, axons project stimulus toward the primary auditory cortex
-----pg.15/pg.16
in an
array called acoustic radiation. (Bear et al
2001: 355-372) The primary auditory cortex
registers the incoming sound images on tonotopic maps. (Bear et al 2001: 380-381) These maps are probably composed of strips of
neurons, called isofrequency bands, each band
handling fairly similar characteristic frequencies. “In
addition to the frequency tuning that occurs in most cells, some neurons are intensity
tuned, giving a peak response to a particular sound intensity.” (Bear et al 2001: 381) Some
neurons even key in on clicks, bursts of noise, frequency-modulated sounds and animal
vocalizations.
With
respect to linguistic communication, there is unanimity in assertions that “specialized language areas have evolved in the human brain that endow us with
an incredibly flexible and creative system for communication.” (Bear
et al 2001: 673) Deacon states, “Without question, children enter the world predisposed to learn human
languages,” (1997: 102) and cites neurobiological,
anthropological and clinical evidence to suggest that the brain has been “significantly overbuilt for learning symbolic associations.” (1997: 413)
These observations notwithstanding, there is also agreement that linguistic processing should not be defined narrowly in terms of spoken language comprehension and production, but communication facility within a broad conceptual system. (Lamb 1999: 238) For instance, in some cases in which American Sign Language users are impaired in a way analogous to Broca’s Aphasia “…the ability to move the hands is not impaired (i.e., the problem is not with motor control); rather, the deficit is specific to the use of hand movements for the expression of language.” (Bear et al 2001: 650) This demonstrates that linguistic communication is possible in the total absence of the sound images normally associated with linguistic processing, even though areas of the
-----pg.16/pg.17
brain
known to facilitate spoken language are in use. Such
flexibility is built into the system by the ways that individual neurons act and interact: “Each brain is formed
in such a way that its wiring and dynamics are enormously variable at the level of its
synapses. It is a selectional system, and
each brain is therefore unique.” (Edelman and Tononi 2000: 213)
There is an unfortunate
misunderstanding among linguists that neuroscientists commonly disagree with one another
at the most basic levels. In fact, as we have
seen in the preceding section, neurobiological researchers are in agreement on a wide
range of issues. It is time to address the
questions posed earlier with respect to neurobiology.
1. Is there a critical period in
language learning?
From
a neurological perspective, there most definitely is a critical period in language
learning. Without neuronal branching, the
neural system would not exist and without selection and entrenchment, the neural system
would be ill equipped to do anything practical. Furthermore,
the number of synapses a neuron is capable of sustaining has a limit called synaptic capacity, which decreases as a neuron
matures. Bear et al note that in the striate
cortex, “the synaptic capacity of immature neurons exceeds
adult cells by about 50%,” and that (in macaque monkeys, which
are similar to humans in many respects) “synaptic capacity in
the striate cortex was remarkably constant from infancy until puberty,” after which synaptic capacity declined sharply. (2001: 721)
-----pg.17/pg.18
2. If
there is a critical period, how does it work?
Critical
periods can be genetically explained by the topobiological
nature of gene differentiation, (Edelman 1992: 58-63) and Edelman notes that for
language, a critical period would likely be “related to
extensive synaptic and neuronal group selection occurring up to adolescence, after which
time such changes occur much less extensively and in a different fashion.” (1992: 130) Remarking on the
advantages of the relatively slow neural development seen in humans as compared to other
species, Deacon observes that “Immaturity of the brain is a
learning handicap that greatly aids language acquisition.” (Deacon
1997: 141)
Although
critical periods do end, neural plasticity does not ever completely disappear: “the environment must modify the brain throughout life at some level, or there
would be no basis for memory information.” (Bear et al 2001:
737) About general neural plasticity, Bear
et al observe, “Early in development, gross rearrangement of
axonal arbors is possible, while in the adult, plasticity appears to be restricted to
local changes in synaptic efficacy.” (2001: 736) While some linguists have asserted that the idea
of a critical period “clearly implicat[es] an innate mechanism”
for language learning, (Eubank and Gregg 1995: 39) this need not be the
case. Critical periods are at least
partially motivated by the general topobiological framework mentioned above. Neuroscientists can account for critical periods
straightforwardly in neurobiological terms; perhaps linguists should spend less time
generating highly speculative critical period theories to match equally speculative
linguistic hypotheses, and more time correlating the body of empirical findings in applied
linguistics with neurobiological critical period knowledge already available.
-----pg.18/pg.19
3.
Is there biological evidence for a Language Acquisition Device that
renders explicit error correction irrelevant?
All
of the neuroscientists mentioned agreed that the brain, in its entirety, is an excellent
language acquisition device, but this fails to respond to the heart of the question. None of the researchers mentioned any candidate
structures that could be construed as a LAD. On
the contrary, all stressed that phrenology (an emphasis on discrete localized function) is
not an option, and that different cognitive skills are distributed throughout the brain’s various systems. Lamb sums
things up: “there is no reason for supposing that the brain
has genetically built-in structures dedicated specifically to language.” (1999: 371) Although genetically
predetermined structures specific to language are not ruled out, until neurobiological
proof affords itself, there is no necessity to posit them.
Finally, aside from the decided lack of biological evidence for a LAD, Deacon
argues that genetic specification of precise neural connections in the brain would be too
costly in terms of genetic resources. (1997: 197)
In fact, learning in the absence of explicit error correction may be explained without hypothesizing an innate LAD. Terry Regier, in his constrained connectionist modeling of basic spatio-linguistic understanding, has offered a cogent explanation. (Regier 1996) As children learn their first words of language, for each word learned, they may understand an implicit negation of that word’s meaning for all other words. While such an oversimplistic view of language does not reflect the way that the meaning of a word tends to overlap somewhat with the meanings of similar words, it matches well with Deacon’s comment concerning neural development that nature tends to overproduce and trim to fit. As Regier's connectionist model takes cues from the actual
-----pg.19/pg.20
systems
of human perception for biological motivation, the high correlation between his model and
actual human language understanding cannot strictly be considered incidental.
4.
Is the brain modular in its processing of language?
The
modules that Pinker suggests (namely, phonology, lexicon, morphology, syntax and
semantics) will not likely be found in any clearly demarcated form because of the
distributed nature of neural processing. Making
a similar observation, Deacon notes that while researchers have made claims for
grammatical modularity by citing studies of Broca’s aphasia in
English speaking patients, the results become much less clear when studying the speech of
Broca’s aphasics whose native language is Italian (a highly
inflected language). (1997: 307) His conclusion:
So if there is a grammar module, then the parts of this module map in very different ways to different grammatical operations, depending on the relative importance of positional or inflectional tricks for cuing grammatical decisions in different languages. This sort of module is a will-o’-the-wisp. (Deacon 1997: 307)
Lamb
echoes this opinion, noting that the linguistic system is not a unified system, but a “complex of subsystems” that are “closely related to one another,” although these
relationships are not simple ones. (1999: 37)
The most likely candidate for modularity would seem to be phonology, in that the primary auditory cortex does seem to be adapted specifically for the reception of sound images into tonotopic maps.
-----pg.20/pg.21
Unfortunately
for a language module theory, it processes all kind of sounds, not just language, so it
could not strictly be called a phonology module. The
other great facilitator of spoken language processing, Wernicke’s area, often seems to play a role among deaf people that is highly analogous
to the role it plays among the hearing. (Bear et al 2001: 650) Can a “phonology
module” that is also capable of processing non-phonological
sign language accurately be called a phonology module?
As will become apparent in question 5, modules for lexicon, morphology, syntax and
semantics are also highly unlikely.
5.
Are the terms associated with modularity biologically meaningful?
At
first glance, different areas of the brain seem to facilitate certain cognitive functions,
but do the functions fall along traditional linguistic lines? Lamb remarks, “The
fact that [patterns of analytical linguistics] can be found in the products of mind doesn’t
necessarily mean that they are direct reflections of anything in that mind.” (1999:
229) Although the brain is very good at language, it is also good at things other than
language. Specifically, Deacon points out:
Though breaking up language analytically into such complementary domains as syntax and semantics, noun and verb, production and comprehension, can provide useful categories for the linguist, and breaking it up according to sensory and motor functions seems easier from a global neuronal viewpoint, we should not expect the brain’s handling of language to follow either of these categorical distinctions. (Deacon 1997: 298)
Edelman has a similar skepticism about the modularity of syntax,
-----pg.21/pg.22
in
particular, although for a different reason: he sees general cognition as a framework for
semantics. He views syntax as a subsequent
epigenetic phenomenon that occurs as rules “developing from
memory” are “treated as objects of
manipulation.” (1992: 130)
Lamb mentions that analytical linguistics, because of its specific formal agenda,
partitions language with respect to “taxonomic convenience”
rather than functional precision. (1999: 32) Despite the tendencies for
Deacon and Edelman to downplay semantics in favor of more general neural function, both
scholars still seem to understand semantics as a driving force in language. (Deacon 1997:
135-136; Edelman 1992: 130) Lamb explains
that the idea of lexical connections is biologically meaningful, “not because they have meanings,
but because they have connections” with the conceptual. (1999: 122)
6.
Is there any neurobiological evidence for either the unity or
separation of production and perception?
In fact, neurobiologically speaking, the two aspects do seem to be somewhat separate, although there must also be some overlap. Incoming sound requires preprocessing in preliminary auditory cortices before moving through the primary auditory cortex to Wernicke’s area in the brain. In contrast, phonetic coding of meaning for speech begins with association in regions including Broca’s area, thereafter proceeding to various motor cortices for the initiation of motor sequences that trigger physical movement of the lungs, vocal chords, tongue and mouth. To the extent that Wernicke’s area and Broca’s area stimulate the same conceptual neural maps, there will be a correlation of activity, but there is a great functional difference between a sound image to be interpreted and a motor schema to produce a sound through speech.
-----pg.22/pg.23
Regarding
this neurobiological dissociation of early sensory and motor cortices, Lamb notes, “…understanding is perceptual while producing language is a motor activity,
like drawing a picture or dancing. If this is
so, language has both a production subsystem and perceptual subsystem as separate systems.”
(1999: 126)
Because
the act of saying a word produces sound that can then be heard by the ear, speech will
result in a reciprocal sound image within the speaker’s ear. Consequently, a strong synchronous firing is
likely to link the neural patterns of the sound image and the motor sequence, even though
the respective maps may be located in separate areas of the brain. (Lamb 1999: 271) Since most people can easily repeat words they’ve just heard, we understand that these correlative connections have indeed
been made. This does not change the fact that
the actual functions are not two sides to the same coin; they are two different coins, or
more accurately a coin and a machine that can mint a similar coin.
7. Does neurobiological evidence show any “economic” necessity to break words down into
syllables or segments?
The assumption that it is uneconomical for the brain to store different forms of words intact assumes that the brain is strapped for memory resources. Is this an accurate assessment? Edelman estimates “there are about 1 million billion connections in the cortical sheet,” and, in taking into account how these connections might be combined, the number of potential connections would be “on the order of ten followed by millions of zeros” (1992: 17) Damasio’s more conservative estimate notes that one human brain contains several billion neurons with at least 10 trillion synapses among these. (1994: 259) For Deacon,
-----pg.23/pg.24
however,
raw neural capacity is not the decisive issue in answering questions of “economy”:
Time is a critically important factor, especially in an information processing device that tends to operate almost entirely in parallel (instead of funneling all operations through a single processing unit, as do most desktop computers) […] Maintaining a signal within a circuit long enough to analyze its part in some extended pattern would tend to get in the way of processes that require rapid and precise timing. (Deacon 1997: 292-293)
Lamb
goes so far as to speculate on the availability of neural wiring over a lifetime, once for
a “maximally curious and energetic” person (in terms of overall cognition), and again with reference to language
for the expected lifetime neural demands of a 20 language polyglot (1999: 341-343); after
auditing the neural balance sheets, he echoes Deacon’s
optimism concerning the brain’s ample resources, stating
simply, “the abundance hypothesis seems to be confirmed.”
(1999: 343) The main neural constraint on language would appear to be
processing time rather than memory capacity.
8. Wouldn’t it be
possible for second language learners to use conventionalized “chunks”
for real communication even lacking combinatorial analysis?
Going beyond Bialystok’s guarded reference to linguistic “chunks” or “meaning units,” there is strong evidence that some aspects of language become deeply entrenched by frequent repetition in production. These “overlearned” linguistic items do indeed seem to operate
-----pg.24/pg.25
irrespective
of grammatical analysis. Bear et al note that
“…there are certain “overlearned”
things Broca’s aphasics can say without
much hesitation, such as the days of the week.” (2001: 643)
While Bialystok and others treat this sort of example as an atypical process that may or
may not have linguistic value, there is no neurobiological reason to believe that all
language is not learned in “chunks” of varying length and grammatical complexity.
In fact, Edelman asserts: “to build syntax or the bases
for grammar, the brain must have reentrant structures that allow semantics to emerge first (prior to syntax) by relating phonological
symbols to concepts.” (1992: 130) Deacon explains:
Indeed,
grammatical cues, such as are embodied in small “function
words,” may be the primary agents for initially tagging and
distributing sentence “chunks” to
be separately processed. For this reason, it
is precisely these features of language that need to be subject to minimal symbolic
analysis. They serve a predominately indexical function.
And as we have seen, indices can be interpreted in isolation as automated,
rote-learned skills. (Deacon 1997: 299)
Lamb posits just such an arrangement, characterizing it as “sequence control without constituent structure,” (1999: 255) and gives extensive neurobiological rationale for making such claims. In fact, he explicitly recommends an “exemplar” understanding of the linguistic “chunks” mentioned above. (Lamb 1999: 263) Actually, for construction grammar advocates including William Croft (Croft 1998) and Adele Goldberg (Goldberg 1995), the observation that foreign language students often attempt to learn language by memorizing large “chunks” is
-----pg.25/pg.26
not
in any way problematic; according to their views, it simply shows the way in which
language is actually learned.
9.
Is there any neurobiological explanation for the way in which both
young native speakers and second language learners regress into errors?
It
has already been stressed that human neural wiring is a highly distributed process. Edelman remarks that ongoing behavior of an animal
makes memory “a process of continual recategorization.”
(1992: 102) Deacon
observes, “Learning is, at its base, a function of the
correlations between things, from the synaptic level to the behavioral level.” (1997: 83) Edelman further notes that “The maps
that speak back and forth are massively parallel and have statistical as well as precise
features” (1992: 29) and “perceptual
categorization, which is one of the initial bases of memory, is probabilistic in nature.”
(1992: 194) In this view,
learning is characterized by probabilistic correlation rather than precise specification.
To put it another way, neural mapping, the basic organizing principle of memory, is association of perception according to spatio-temporal contiguity, which is not actually logical at all. This dynamic process, in logical terms, is the fallacy “guilty by association,” but the illogicality in the argument’s form does not stop it from being efficacious. Neurons that fire together, wire together and so logically unrelated aspects of sensory perception can become associated through correlation. Less well-entrenched routines have weaker mutual associations among constituent neurons and therefore a lower statistical probability of being available at the crucial moment. Also, when a neuron fires there is resource depletion; the chemical resources that
-----pg.26/pg.27
allow
for the firing do not remain at constant levels, resulting in further potential
instability for less well-entrenched routines. Just because a language
item is consistently available does not mean it is being accessed in the same way from one
instance to the next.
To summarize section C, there is indeed a critical period in language learning and it is crucially interrelated with neuronal branching and selection processes. The quick pace of language learning in the absence of explicit error correction may be explained according to basic neuronal processes without reference to language modules or a LAD; indeed, there is no neurobiological evidence to support their existence. Consequently, the terminology often associated with modularity, while analytically useful, may not be biologically meaningful. Although perception and production necessarily overlap in association cortices, motor functions of production and sensory functions of perception are relatively discrete. There is no “economic” necessity to break words or phrases down to conserve memory. Not coincidentally,
-----pg.27/pg.28
overlearned
“chunks” of language are
essentially linguistic in nature and an exemplar-based understanding of language is the
most neurally plausible option. Finally,
regression into error for children and second language learners can be explained
straightforwardly in neural terms by the unstable, probabilistic nature of perception and
linguistic association.
In that this type of relevant neurobiological knowledge is readily available, highly theoretical speculation that fails to take known processes into account is unwarranted; so, too, are attempts to mitigate problematic evidence in light of such speculation.
Needless
to say, sections B and C above do not represent the final word on the issues in question,
but taking a neurobiological perspective into account is infinitely better than NOT taking
such a perspective into account.
Skehan, in his book, A Cognitive Approach to Applied Linguistics (Skehan 1998), calls for a realignment of theory with empirical evidence. He mentions a great number of cases in which theory does not match up well with observation. Many of the problems he identifies from a psycholinguistic perspective have been echoed in this paper. Having listed all of these discrepancies between empirical evidence from applied linguistics and mainstream linguistic theory, however, he essentially ignores the mutually exclusive nature of the competing viewpoints by proposing a “dual mode system.” Arguing that both rule-based systems and exemplar-based systems are insufficient, he states: “The question then becomes one of explaining how the two systems might work together harmoniously.” (Skehan 1998: 89) Skehan
-----pg.28/pg.29
reasons
that an exemplar-based system, “with its emphasis on meaning,”
would be too inflexible to accommodate “underlying
system change.” (1998: 89)
But what “underlying” system
is he talking about? In neural terms, we have
already seen that there is no “underlying system,” only largely inextricable subsystems. The
purported difficulty of “underlying system change” is a dilemma only if one has already posited an underlying system. Skehan deftly critiques the theoretical pyramid
only to mitigate his own findings by invoking evidence taken from presuppositions in the
pyramid itself.
Nevertheless,
research aiming for a “cognitive” view
of language acquisition cannot hope to account for everything in terms of neural
processes. Skehan duly notes that the
classroom is a social environment and so affective concerns must also be figured into the
equation. Still, it is unfortunate that
although Skehan’s book is titled A Cognitive Approach to Applied Linguistics, he
makes almost no reference to neurobiology. When
he advocates a cognitive approach, he explicitly refers to a psycholinguistic approach. (Skehan 1998: 2) Although there is a tendency to equate “cognitive” with “psychological,”
the findings of psycholinguistic study represent but one portion of the
potentially useful evidence that may be brought to bear on perennially intractable
problems of applied linguistics. Although
psycholinguistic evidence is important and cannot be brushed aside, ultimately, any theory
of language that is not defensible with respect to neurobiology is not a viable theory.
Theories of language learning are only useful insofar as they detail how language is actually learned and thereby facilitate effective learning.
-----pg.29/pg.30
Theoretical
stances that subsume both the neurobiological underpinnings of language and empirical
evidence from language acquisition research do not serve the goals of applied linguistics. Although knowledge of neurobiology could play a
mediating role in the resolution of longstanding disputes, applied linguists have made
little reference to the actual electrochemical processes by which language is facilitated. A change of perspective may allow applied
linguists freedom to extend the limits of theory, rather than
being restrained by it.
Bear,
M., B. Connors and M. Paradiso. 2001. Neuroscience:
Exploring the Brain, Second Edition. Baltimore,
MD: Lippincott Williams & Wilkins.
Bialystok,
E. 1991.
“Metalinguistic Dimensions of Bilingual Language
Proficiency.” Language Processing in Bilingual Children. (Bialystok, E. ed.) London: Cambridge University Press.
Croft,
W. 1998.
“The Structure of Events and the Structure of
Language.” The New Psychology of Language. (Tomasello, M. ed.) New Jersey: Lawrence
Erlbaum Associates.
Damasio,
A. 1994.
Descartes’ Error:
Emotion, Reason, and the Human Brain. New
York: G.P. Putnam.
Damasio,
A. 1999.
The Feeling of What Happens. New York: Harcourt, Inc.
Deacon,
T. 1998.
The Symbolic Species. New York: W.W. Norton and Co.
Edelman,
G. 1992.
Bright Air, Brilliant Fire: On the
Matter of Mind. New York: Basic Books.
Edelman,
G., and G. Tononi. 2000. A Universe
of Consciousness. New York: Basic Books.
Eubank,
L. and K. Gregg. 1995. “Et in Amygdala Ego?: UG, (S)LA, and Neurobiology.” Studies in Second Language
Acquisition, 17, 35-57. Cambridge:
Cambridge University Press.
Goldberg,
A. 1995.
Constructions: A Construction Grammar
Approach to Argument Structure. Chicago:
University of Chicago Press.
-----pg.30/pg.31
Jacobs,
B. and J. Schumann. 1992. “Language Acquisition
and the Neurosciences: Towards a more Integrative Perspective.” Applied Linguistics, 13. Oxford: Oxford University Press.
Lakoff,
G. and M. Johnson. 1999. Philosophy
in the Flesh. New York: Basic Books.
Lamb,
S. 1999.
Pathways of the Brain. Amsterdam: John Benjamins Publishing.
Pinker,
S. 1994.
The Language Instinct. London: Penguin Books.
Pinker,
S. 1999.
Words and Rules: The Ingredients of
Language. New York: Harper Collins
Perennial.
Regier,
T. 1996.
The Human Semantic Potential. Cambridge, MA: MIT Press.
Schmidt,
R. 1995. “Consciousness
and Foreign Language Learning”
Attention & Awareness in Foreign
Language Learning. (Schmidt, R., ed.) Honolulu:
UH Manoa, Second Language Teaching and Curriculum Center.
Skehan,
P. 1998.
A Cognitive Approach to Language
Learning. Oxford: Oxford University
Press.
Watson,
I. 1991.
“Phonological Processing in Two Languages.” Language Processing in Bilingual Children. (Bialystok, E. ed.) London: Cambridge University Press.
Wexler,
K. and P. Culicover. 1980. Formal
Principles of Language Acquisition. Cambridge,
MA: MIT Press.
Widdowson,
H. 2000.
“On the Limitations of Linguistics Applied.” Applied Linguistics, 21. Oxford: Oxford University Press.