| |
Abstract:
Production errors provide empirical evidence for the
development and testing of models of speech production. This
study employs a corpus of 676 lexical substitution errors to
investigate the interaction of factors relating error words and
their intended targets. Six factors relating error and target
words are analyzed: lexical frequency, syntactic class, word
length in syllables, morphological complexity, word-initial and
non-initial phonological similarity, and semantic relatedness. We
find evidence for (1) the additive nature of spreading activation
in a network of word forms; (2) a two-level model of the lexicon
separating semantics and form; (3) interaction between the two
levels; and (4) inhibitory and editing processes during
production.
Two types of correlations between factors are found,
distributional relationships in the lexicon (e.g., longer words
tend to be less frequent) and compensatory relationships that
arise during lexical processing. The compensatory relationships
discovered are significant negative correlations (in form-based
but not semantic errors) between initial phonological similarity
and non-initial similarity, length, and error frequency.
Compensation is compatible with spreading activation accounts of
access (Dell 1986), and suggest that activation levels in the
network are influenced additively by several factors. The
simplest interpretation is that these factors are organizing
parameters in the form network of the lexicon.
While measures of phonological similarity and semantic
relatedness largely separate the semantic and form-based
substitutions, no compensatory relationship is found between
phonological similarity and semantic relatedness. This supports
models assuming a divide between access of semantics and form
(Dell 1981; Levelt 1989). However, we find a significant
correlation between length and stress patterns of semantic errors
and targets, as well as a greater degree of phonological
similarity in semantic errors than expected by chance. Form
similarity in semantic pairs supports models proposing level
interaction (Dell and O'Seaghdha 1991).
We find a strong correlation between the frequencies of
targets and errors that is independent of other factors
contributing to frequency correlation, and a significant
frequency bias for errors over targets. The results of previous
error bias studies conflict (Levelt 1989; del Viso et al. 1991;
Hotopf 1980; Harley and McAndrew 1995). The bias is here
demonstrated for both form and semantic errors, but is isolated
to polysyllabic words, with a tendency for bias to increase with
length. Frequency correlation suggests either frequency based
lexical organization or a frequency based inhibitory process. The
error bias effect and its relation to syllable length results
perhaps from editing during frame construction.
Dell, G. (1986). "A spreading-activation theory of retrieval in
sentence production."
Psychological Review
93, 293-321.
Dell, G., O'Seaghdha, P. (1991) "Mediated and convergent lexical
priming in language production: A comment on Levelt et al. 1991."
Psychological Review
98, 604-614.
del Viso, S., Igoa, J., Garcia-Albea, J. (1991). "On the
autonomy of phonological encoding: Evidence from slips of the
tongue in Spanish."
Journal of Psycholinguistic Research
20, 161-185.
Levelt, W. (1989).
Speaking: From intention to articulation.
Cambridge, MA:MIT Press.
|
