Brain Imaging & Linguistics

[Colin Harrison]

        I wonder how many of you have read Jaeger et al's piece in Language
(72.451-97) about areas of the brain that are activated during past tense
formation in English.  I understand that it's been quite a hit in some
parts of the linguistic world, and it's certainly to be praised for its
methodological rigor and the honesty of the authors. I am convinced that
this sort of experimentation ought to represent a significant direction for
future research (some of which I hope to be doing soon myself).  The thing
is, the experimenters interpret their results as showing that regular
inflections are processed differently than irregular inflections, but I
don't see that their theoretical conclusions follow from their data due to
at least two major confounds.  I wanted to put these ideas out to the rest
of you Funknetters and see what y'all think.
        First up, semantic discrepancies between Jaeger et al.'s word lists
represent a significant confound.  The two lists of interest are the cue
sets from which subjects had to form regular and irregular past tenses
(sets 3 and 4).  Jaeger et al. note that overall, the irregular past forms
require more cortical activation than the regulars, and conclude that this
is because they are not associated with an on-line rule system, and hence
require more attention and greater resource devotion (p.487).  But if you
look at the meanings involved, a rather different explanation seems at
least plausible.
        Each list comprises 46 tokens.  Of these, the regular past list has
just nineteen (41%) that are unambiguously human physical activities,
involving limb movement.  The irregular list shows a much higher proportion
of human physical activities, 33 of the 46 tokens (73%).  This looks like a
significant difference to me!  Might not the greater cortical activation
noted in the irregular condition be a result of more widespread somatic
activation as an intrinsic part of the meaning of the verbs, rather than
anything to do with their morphosyntactic regularity?  There is ample
evidence emerging from imaging studies (follow up for instance the work of
Hanna and Antonio Damasio), that the comprehension of words that are
connected with any kind of somatic experience involves activation in some
of the same areas as the instantiation of the experience itself.  So, the
meaning of a verb such as "walk" will involve indirect activation of the
somato-sensory circuits necessary to walk, plus all those more peripherally
involved in the experience of the activity etc.
        Jaeger et al's results look as if they represent disconnected
activation patterns, but their results were not so neat and clean at first:
they needed to "wash" a fair amount of "random" noise from their charts
until they arrived at something resembling the neat, discrete pictures they
presented.  They are completely open about the normalising proceedures they
follow, and it's all there in black and white for anyone who wants to
examine it more closely than I have.  My concern is, it's not unlikely that
they could have "washed" out the evidence of simmilar somatic activation
from the regular list, but the somatic activation in the irregular list
would have been too large to remove in this way, leaving behind different
activation patterns based not on algorithmic versus non-algorithmic
processing (Jaeger et al's conclusion), but rather based on the semantic
category of the verbs in each list.
        Secondly, even if we dismiss the first objection, the experimental
design itself assumes the conclusion.  That is, subjects in the test
conditions were performing an algorithmic task at the behest of the
examiner: "given x (a verb stem form), produce y (the past tense of the
same verb)."  It is not clear to me that information about brain activation
during a predictable (and probably pretty boring) two-minute algorithmic
task has any relevance to brain activation during production of similar
forms when one is engaged in meaningful speech.  In order to equate these
two types of processing, you have to begin with the assumption that
speakers inflect verbs according to an algorithmic procedure during on-line
discourse production - exactly the kind of process whose centrality to
natural language production is disputed!

What do you think?


Colin Harrison
Dept. of Linguistics
Rice University
Houston TX 77030
USA