|
Abstract:
Tabor, Juliano & Tanenhaus (TJT, 1997) presented a series
of experiments that they interpret as demonstrating that the human
sentence processor is keeping track of contingent frequencies in
ambiguity resolution. TJT showed 1) that the preferred resolution
of the English complementizer/determiner ambiguity of the word
"that" is towards the determiner sentence-initially, but towards
the complementizer following a verb, and 2) that the post-verbal
preference for the complementizer reading holds even when the
subcategorization of the verb makes this interpretation impossible:
in (1) "that" is read more slowly than the control "those".
(1) The women visited that/those beautiful garden(s) to see the
wide variety of roses.
TJT hypothesized from this second finding that the human
language processor treats contexts following verbs similarly.
An alternative to TJT's account of these results is a
combination lexical (unigram) frequency/structural complexity
approach, like that of Gibson (1998). The word "that" is much more
frequent as a complementizer than as a determiner overall
(independent of context), so lexical frequency always favors the
complementizer reading in this approach. Post-verbally the
complementizer is favored because of this lexical frequency;
structural complexity has no preference: Two syntactic heads (a
noun and a verb) are required to complete the input as a sentence
under both the complementizer and determiner readings.
Sentence-initially, however, structural complexity favors the
determiner reading, because only two syntactic heads are required
to form a sentence from a sentence-initial determiner (a noun and a
verb), versus three from a sentence-initial complementizer (a noun
and two verbs). If the structural complexity preference outweighs
the lexical frequency preference, the observed results follow.
Contrary to TJT, this approach predicts the word "that" should
be processed slowly relative to "those" in all contexts, not just
post-verbally. Two self-paced word-by-word reading experiments were
conducted to test this prediction. Experiment 1 crossed
determiner-type (that, those) and preceding-category (verb,
preposition), as in (1) and (2):
(2) The women in that/those beautiful garden(s) will soon be
photographed.
TJT's hypothesis predicts an interaction, with the "that"
condition being slower than the "those" condition following a verb
but not following a preposition. The results were as predicted by
the lexical frequency approach: the "that" conditions were read
more slowly than the "those" conditions in both verbal and
prepositional environments (60 subjects, 20 items, p<.05 by
subjects and items), with no suggestion of an interaction.
Experiment 2 compared "that" and "those" sentence-initially:
(3) That/Those experienced diplomat(s) would be very helpful to
the lawyer.
TJT's contingent frequency approach predicts no difference
between the "that" and "those" conditions here. The results
supported the lexical frequency approach: the "that" condition was
significantly slower than the "those" condition immediately on the
determiner (32 subjects, 24 items, p<.01 by subjects and
items).
We therefore conclude that the difference between "that" and
"those" initially observed by TJT is a lexical frequency effect
independent of context. Furthermore, because TJT's evidence was the
strongest source of evidence in favor of models which track
contingent frequencies, the hypothesis that the parser keeps track
of such frequencies needs to be re-evaluated.
|
