The influence of semantics on past tense inflection

Abstract

The influence of semantics on past-tense inflection Michael Ramscar (michael@cogsci.ed.ac.uk) University of Edinburgh, 2 Buccleuch Place Edinburgh, EH8 9LW, Scotland. Abstract Semantics and past-tense inflection Previous theories of past-tense verb inflection have considered phonological and grammatical information to be the only relevant factors in the inflection process (e.g. Bybee & Moder, 1983; Rumelhart & McClelland, 1986; Kim, Pinker, Prince & Prasada, 1991). This paper presents three experiments that show that semantic information plays a decisive role in determining the inflection of both existing and novel homophone verb stems. These findings indicate that regular and irregular inflections are determined by semantic and phonological similarities in memory, and furthermore that people are not responsive to the kind of grammatical distinctions amongst verb roots that default rule theories of inflection (Pinker, 1999) presuppose. Introduction In most theories -- and studies -- of past-tense verb inflection, phonological and grammatical information have been considered to be the two relevant factors in the inflection process (e.g. Bybee & Moder, 1983; Rumelhart & McClelland, 1986; Kim, Pinker, Prince & Prasada, 1991; Pinker, 1991; 1999). However, in some models of inflectional processing (MacWhinney & Leinbach, 1991; Joanisse & Seidenberg, 1999), semantic processes have been included to help explain the processing of homophone verbs (e.g. brake/break). Since brake and break both sound the same, phonology alone cannot distinguish which of broke or braked is to be the correct past tense form for the input bre I k . Although using semantic information to guide this process appears intuitively plausible, it has not been supported empirically, and indeed this suggestion has been fiercely criticised by Pinker and colleagues (Kim et al, 1991; Pinker, 1999), who put forward an alternative, natavist account of homophone inflection (Pinker, 1991; 1999). This predicts that the regularisation of irregular sounding verb stems is driven by innate grammatical sensitivity: verbs that are instinctively perceived to be denominal will be automatically regularised. This account is supported by results reported by Kim et al (1991) which indicate that grammatical factors correlate better than semantic factors with people's ratings of the acceptability of past tense forms in context, although these results do not rule out any semantic role in inflection. So do semantics have any influence on the past tense forms speakers produce? This paper seeks to clarify and directly address this question. To initially test whether semantic similarity can affect the inflection of verb past tenses, Experiment 1 examines the past tense forms native English speakers produce for novel (nonce) English verbs whilst holding phonological factors constant and varying the semantic contexts in which the verbs are presented. The phonologically similar nonces sprink and frink are presented in contexts that primed either the existing phonologically similar regular forms blink or wink, or the existing phonologically similar irregular form drink. It is hypothesized that if semantic similarity played a part in the inflection process, then there will be significant differences between the proportion of regular and irregular forms produced, in line with whether the semantic context favored an existing regular or irregular verb. Experiment 1 Participants. Participants were 120 visitors a shopping mall in Edinburgh, Scotland, and 40 students at the University of Edinburgh. All were native English speakers and participated voluntarily in the study Materials. A set of cards were printed with a paragraph that contained a highlighted nonce verb (sprink or frink) in a context in which the nonce was in its infinitive tense, and a blank that later required its past tense. Two of the contexts were further manipulated so that they primed either two existing regular verbs -- blink and wink -- that are phonologically similar to the nonces, or an existing irregular verb -- drink -- that is phonologically similar to the nonces. The contexts constructed are shown in Table 1: in the drink context, the nonce was shown in a context that used it as a verb to describe the consumption of vodka and fish, whereas in the blink and wink context the nonce was shown in a context that used it as a verb to describe a symptomatic affliction of the eye-lid associated with a fictitious disease. A third context was designed to semantically prime neither drink nor blink or wink (instead the nonce was used as a verb describing a hypnotic trance and was semantically similar to the regular verb meditate; see Table 1), whilst a control presented the nonce in a context that provided few semantic clues ( John likes to frink. Last week he In order to control the phonological properties of the nonces in the semantic contexts, both the initial presentation of the nonce, and the blank which was used to elicit the past tense form from participants were embedded in the same sentence substructure in each of the three