The /s/ morpheme and the compounding phenomenon in English

Abstract

The /s/ morpheme and the compounding phenomenon in English Jenny Hayes 1 (J.Hayes@herts.ac.uk) Victoria Murphy 1 (V.A. Murphy@herts.ac.uk) Neil Davey 2 (N.Davey@herts.ac.uk) Pamela Smith 1 (P.M.Smith@herts.ac.uk) Lorna Peters 2 (L.Peters@herts.ac.uk) Departments of Psychology 1 and Computer Science 2 , University of Hertfordshire, College Lane, Hatfield, AL10 9AB, United Kingdom. Abstract Compound words with irregular plural nouns in first position (e.g. mice-eater) are produced far more frequently than compound words with regular plural nouns in first position (e.g. *rats-eater), (Gordon, 1985). Using empirical evidence and neural net modelling, the studies presented here demonstrate how a single route, associative memory based account might provide an equally, if not more, valid explanation of this phenomenon than the standard dual mechanism based theory (Marcus, Brinkmann, Clahsen, Weise & Pinker, 1995). 1. Introduction 1.1 The Compounding Phenomenon Psycholinguistic research has shown that English compound words with irregular plural nouns in first position (e.g. mice-eater) are produced far more frequently than compound words with regular plural nouns in first position (e.g. *rats-eater), (Gordon, 1985). 1.2 The Dual Mechanism Model’s Explanation of Compounding The dual mechanism model (Pinker, 1991 ), proposes that irregular nouns and their plurals are stored as memorised pairs of words in the mental lexicon (e.g. mouse-mice) but that regular plurals are produced by the addition of the /s/ morpheme to the regular stem at a post lexical stage (e.g. rat + s = rats). Compounds are created in the lexicon. Thus as irregular plurals are stored in the lexicon they are available to be included within compound words. However, as only the singular stems of regular nouns are stored in the lexicon the plural form is never available to be included within compound words (Marcus et al, 1995). 1.3 A Single Route Associative Memory Based Explanation of Compounding An alternative explanation of this compounding phenomenon based on the frequency and patterns of occurrence of items in the linguistic input has not been explored fully. However an explanation of this sort may explain the treatment of both regular and irregular plurals in compounds (Murphy, 2000). Frequency counts of a sample of the CHILDES (Child Language Data Exchange System) corpora (McWhinney & Snow, 1985) have shown that the plural /s/ morpheme is a perfect predictor of word finality and furthermore, the plural /s/ morpheme is never followed by a second noun. Importantly, the reverse pattern is found with the possessive /‘s/ morpheme since it is always followed by a second noun. Therefore, it might be that a noun rarely follows the regular plural /s/ morpheme (i.e. patterns such as “*rat/s/ chaser” do not occur ) because the pattern “noun – morpheme /s/- noun” is reserved for marking possession (such as rat’s tail). Interestingly in other languages that do not have this competition between the plural and possessive morpheme such as Dutch (Schreuder, Neijt, van der Weide & Baayen, 1998) and French (Murphy, 2000), regular plurals are allowed within compounds. Irregular plurals may, however, appear in English compounds as they are not formed by the addition of the plural /s/ morpheme. Thus, irregulars do not compete with the possessive structure and as such may be followed by a second noun in a compound. This polyfunctionality of the /s/ hypothesis is explored here using three neural net simulations and an empirical study. 2. Neural Net Modeling An associative memory-based account of inflectional morphology has been investigated in numerous connectionist models. Several models have successfully simulated the putative dissociation between regular and irregular inflection for both verbal morphology (Daugherty & Seidenberg, 1994) and plural morphology (Plunkett & Juola, 1999) using a single learning mechanism and no explicit rules. Furthermore, as well as being able to learn mappings from input to output, connectionist models have also been able to learn sequential mappings (Elman 1990). Thus it is predicted that a single route associative memory system could learn that the inclusion or omission of the regular plural morpheme /s/ is influenced by where that /s/ morpheme occurs in a sequence of language input. Three neural net