On the interpretation of frequency effects in comprehension and production.

Add to your list(s) Download to your calendar using vCal

• Harald Baayen (MPI for Psycholinguistics, The Netherlands)
• Wednesday 29 November 2006, 16:00-17:30
• GR-06/07, English Faculty Building.

If you have a question about this talk, please contact Susan Rolfe.

Harald Baayen, Laura Balling, Wieke Tabak, Heidrun Bien and Hanke van Buren Radboud University Nijmegen and MPI for Psycholinguistics, The Netherlands

Ever since Taft (1979), the word frequency effect has been used as a tool to gauge the representation and processing of morphologically complex words. Two frequency counts have received special attention: the frequency of the complex word as a whole (henceforth Complex Frequency), and the frequency of its stem (henceforth Stem Frequency). Stem Frequency effects have been advanced as the hallmark of constituent driven processing in procedural memory, while Complex Frequency effects have been interpreted as evidence for item specific information available in declarative memory (Pinker & Ullman, 2002).

In this paper, we argue that Complex Frequencies do not reflect the activation of or access to item specific representations (similar to those of simple words). Instead, we think that they reflect memory traces of previous combinatorial processing in procedural memory. We also argue that Stem Frequency is an inappropriate measure for gauging constituent based processing. Instead, Stem Frequency seems to tap into a word’s conceptual familiarity.

Our arguments are based on two strands of evidence. The first strand concerns the significance of Complex Frequency and the nonsignificance of Stem Frequency as predictors of lexical decision latencies in regression studies of visual and auditory comprehension of relatively low frequency words in English, Spanish, and Danish. The traditional interpretation predicts Stem Frequency effects rather than Complex Frequency effects for such words, the mirror image of what we actually observe. Interestingly, the absence of a significant Stem Frequency effect is paired with the presence of various other predictors indicating constituent driven processing. This leads to the conclusion that for comprehension, Stem Frequency (as a simple unigram probability) is not the right measure for gauging parsing processes, and to the inference that Complex Frequency (as a joint probability) is a measure of experience with morphological processing.

A second strand of evidence concerns speech production. According to the WEAVER (Levelt, Roelofs, Meyer, 1999) model, the Stem Frequency effect would arise during the encoding of a word’s form. We used picture naming experiments to probe the early stages of speech production, and position response naming to investigate the later stages of word form encoding and articulatory preparation. For inflected words, we observed a U shaped effect of Stem Frequency for picture naming, but not for position response naming. This suggests that Stem Frequency effects arise during conceptualization and perhaps lemma selection, without, however, playing a role during later stages of form encoding. Instead, word form encoding seems to be subject to lexical competition, as gauged by measures of phonological density and cohort entropy.

Complex Frequency was never a good predictor in our production experiments, as predicted by decompositional models such as WEAVER . However, measures gauging the complexity of a word’s inflectional paradigm either a type count of inflectional variants (for Spanish), or Shannon’s entropy calculated over these inflectional variants (for Dutch) were predictive, both in picture naming and in position response naming. This suggests that throughout the production process, the selection of the appropriate inflected form from its inflectional paradigm is a computationally demanding probabilistic process that is very different from straightforward selection schemas such as implemented in WEAVER . The effects of inflectional entropy also bear witness, albeit indirectly, to the importance of the probabilities of words’ inflectional variants, contrary to what fully decompositional models would suggest.

Considered jointly, our results caution against linking frequency effects straightforwardly to hypothesized mental representations of words and their constituents, and call for more sophisticated distributional measures.

This talk is part of the RCEAL occasional seminars series.

This talk is included in these lists:

• GR-06/07, English Faculty Building
• RCEAL occasional seminars

Note that ex-directory lists are not shown.