Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Normal and Impaired Processing in Quasi-Regular Domains of Language: The Case of English Past-Tense Verbs

Karalyn Patterson (1), Matthew A. Lambon Ralph (1), Helen Bird (1), John R. Hodges (2), James L. McClelland (2)

(1) MRC Cognition & Brain Sciences Unit, Cambridge, UK
(2) CNBC, Carnegie Mellon University, Pittsburgh, PA, USA 11122

In most if not all languages, there are domains in which the relationship between one form of a word and another can be described as quasi-regular [1]. This means that, across the whole relevant vocabulary, there is substantial but imperfect consistency in the nature of the transformation linking the two forms. Quasi-regularity may apply between the same word forms in different modalities, as in the spoken and written versions of words, and also between different morphological forms of the same word root. In many languages, some aspects of verb morphology have this feature. The majority of English verbs, for example, are transformed to past tense by simply adding -ed to the present tense form (e.g., talk -> talked). The remainder, which include some of the most commonly used verbs in the language, form their past tenses in an atypical way (e.g., speak -> spoke; think -> thought; have -> had).

According to one widely held theory about language, the nature of processing for regular (and novel) vs. irregular transformations differs so fundamentally as to require two completely separate mechanisms. In the realm of the English past-tense, this theory has been most thoroughly described by Pinker [2]. An alternative view, most recently developed by Joanisse & Seidenberg [3], proposes that all types of past-tense transformation are achieved by a single distributed, constraint-satisfaction process recruiting activation of the phonological and semantic representations of words.

Among various forms of empirical evidence pertinent to this theoretical debate, attention has recently turned to the performance of neurological patients with acquired language impairments. One popular paradigm assesses generation of the past-tense in a sentence frame preceded by the present tense of the same verb (┘coday I speak to my friend; yesterday I ___ to my friendíĘ). Previous evidence is reviewed and new evidence presented here for a double dissociation in this task, with some types of aphasic patients achieving greater success on regular and novel verbs whereas others show an advantage for irregular verbs. Although the dual-mechanism theory predicts this double dissociation, such a pattern of results can also be well explained by a single constraint-satisfaction process. The crucial proposal here is that disruption to semantic representations has a disproportionate impact on the processing of irregular forms, and that disruption to phonological processing is more detrimental to success with regular forms. The semantic and phonological capabilities of patients with the two sides of the dissociation therefore become salient evidence in this debate. Additional features of the evidence, such as the nature of the patients' errors, seem to favour the constraint-satisfaction approach.

References

  1. Plaut, D.C., McClelland, J.L. Seidenberg, M.S., Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56-115.
  2. Pinker, S. (1999). Words and Rules: The Ingredients of Language. London: Weidenfeld & Nicolson.
  3. Joanisse, M.F. & Seidenberg, M.S.. (1999). Impairments in verb morphology after brain injury: A connectionist model. Proceedings of the National Academy of Sciences, 96, 7592-7597.


Full Paper

Bibliographic reference.  Patterson, Karalyn / Ralph, Matthew A. Lambon / Bird, Helen / Hodges, John R. / McClelland, James L. (2000): "Normal and impaired processing in quasi-regular domains of language: the case of English past-tense verbs", In ICSLP-2000, vol.2, 15-19.