| |
Abstract:
We found that imageability ratings of irregular but not
regular English verbs correlated with their past-tense
acceptability ratings. We argue that the data suggest that
irregular but not regular past-tense forms are retrieved from
memory.
Competing theories of the computational basis of language have
recently focused on regular and irregular morphology. One view
posits that all but the most arbitrary morphological
transformations are computed by mental rules (Halle & Mohanan,
1985). A second posits that rules are only descriptive entities,
and that all transformations are computed in an associative memory
that can be modeled by connectionist networks (Rumelhart &
McClelland, 1986; Joanisse & Seidenberg, 1998). A third
proposes a hybrid dual-system model, with regulars computed in
real-time by the application of mental rules, and irregulars
dependent upon associative memory (Pinker, 1991).
Words whose meanings are highly concrete, or are easily imaged,
are better remembered than those that are less concrete or
imageable (Paivio, 1995). If past-tense forms are retrieved from
memory, their computation should be sensitive to verb imageability:
Past-tenses of verbs that are easily imaged should be better
remembered than those with low imageability. If instead past-tenses
are produced in real-time by a mental rule, they should show no
such imageability effects, once access to their stem forms is held
constant.
Native English-speaking adults gave stem familiarity and
past-tense acceptability ratings, and verb imageability ratings, to
monosyllabic regular and irregular verbs. Each past-tense form was
preceded by its stem, precluding the necessity to retrieve the stem
from memory in order to compute its past-tense.
Imageability ratings correlated with past-tense ratings of
irregulars (p<.05) but not regulars (p>.10), partialing out
past-tense frequency. Imageability ratings correlated with stem
ratings of both verb types (p<.05), partialing out stem
frequency - as expected, since the stems must be stored.
The findings suggest that the computation of irregular but not
regular past tense is dependent upon lexical-conceptual memory
(Pinker, 1991; Ullman et al., 1997). The results are incompatible
with connectionist models with no semantic representations
(Rumelhart & McClelland, 1986), and pose a challenge to models
that posit the real-time rule-based computation of irregular forms
(Halle & Mohanan, 1985). The data may be inconsistent with
connectionist models with distinct representations for semantics
and phonology (Joanisse & Seidenberg, 1998; Plaut et al.,
1996), because both verb types require semantic representations,
and therefore they might be expected to show similar imageability
effects. Finally, the findings demonstrate that conceptual
measures, and imageability in particular, can be used as an
investigative tool for probing the computational basis of
linguistic forms.
References
Halle, M., & Mohanan, K. P. (1985). Segmental phonology of
modern English.
Linguistic Inquiry,
16(1), 57-116.
Joanisse, M. F., & Seidenberg, M. S. (1998). Dissociations
between rule-governed forms and exceptions: A connectionist
account. University of Southern California.
Paivio, A., (1995). Imagery and memory. In M. S. Gazzaniga (Ed.),
The Cognitive Neurosciences,
pp. 977-986. Cambridge, MA: MIT Press.
Pinker, S. (1991). Rules of language.
Science,
253, 530-535.
Plaut, D. C., et al. (1996). Understanding normal and impaired
word reading: Computational principles in quasi-regular domains.
Psychological Review,
103(1), 56-115.
Rumelhart, D. E., & McClelland, J. L. (1986). On learning the
past tenses of English verbs. In J. L. McClelland, D. E. Rumelhart,
& PDP Research Group (Eds.),
Parallel Distributed Processing: Explorations in the
Microstructures of Cognition,
pp. 216-271. Cambridge, MA: Bradford/MIT Press.
|
