The Morphology of the Mental Lexicon: Internal Word Structure Viewed from a Psycholinguistic Perspective
Dominiek Sandra
A conceptual analysis is made of several ways in which the morphological structure of words might enter their lexical representation and/or processing. Economising on storage space seems an attractive option in light of the linguistic definition of the morpheme. However, for several word types, problems would arise in the mapping of morphs onto morphemes and of morphemic meanings onto whole-word meanings. Moreover, economy is a legal option, which might not be available to the mental lexicon. Alternatively, morphs might be put to the purpose of increasing lexical access speed, as proposed by Taft and Forster (1975) in their prefix-stripping model. It is demonstrated that such a view strips morphology from virtually all of its linguistic aspects. Furthermore, the prefix-stripping model would decrease rather than increase the access speed for several types of prefixed words in a number of languages. Linguistically interesting hypotheses are instantiated by the view that affixes are used for providing the syntactic processor with structural information and by the proposal that morphologically organised representational structures are the result of an encoding/retrieval scheme developed at the time of lexical acquisition/learning. Finally, morphemic units might be involved in processing as the result of the frequency of their letter cluster. It is a remarkable fact in current psycholinguistic research on morphology that, despite the linguistic nature of the study object, the majority of studies addressing it have been concerned with access representations rather than intra-lexical, linguistically relevant representations.
INTRODUCTION
As morphemes are considered to be the structural units of words, psycholinguists studying the word recognition process sooner or later had to venture into the linguistic domain of morphology. Indeed, some initial explorations by Gibson and Guinet (1971) and Murrell and Morton (1974) in the early 1970s were soon followed by the seminal research of Taft and Forster (1975; 1976). This work led to a steady interest in the topic and generated a large literature.
There are two simple and easily observable facts about morphology that make this part of linguistic structure an intriguing area for psychological investigation. First, it is striking to observe large numbers of polymorphemic words in the lexicon of numerous languages of the world (if not all of them). Second, the fact that language users coin and understand novel polymorphemic words rather frequently and without any noticeable difficulty attests to the importance of morphology in language use. This pervasiveness of morphological structure, both in the word stock of languages and in speakers' lexical creativity, suggests that the morphological dimension in language is not a linguistic accident but serves an important purpose in language use. Moreover, it suggests that the lexical processing system of language users is fairly well designed for dealing with polymorphemic words.
These facts entail a number of psycholinguistically interesting questions. What is the nature of the processes involved in producing and understanding novel polymorphemic words? What insights can be gained from the psycholinguistic study of existing polymorphemic words? Do language users draw on their perception of morphological relations between words for storing these words in their mental lexicon? If so, how does this translate into representational language, that is, what do the representations of polymorphemic words look like? Furthermore, what are the precise consequences for lexical processing? More particularly, at what processing level is morphological structure involved and what purpose does it serve there? Does the mental lexicon treat polymorphemic words as a monolithic class or does it accommodate different word types in different ways? If the latter possibility is correct, which linguistic variables are relevant for defining those word types? Is it possible to generalise across languages or are language-specific properties of morphological structure important determinants of the way polymorphemic words are represented and processed? How do the properties of the various language modalitiesâreading, writing, listening and speakingâaffect the answers to the above questions?
A psycholinguistic theory on morphological involvement in word processing can be complete only if it provides an answer to all the above questions. Of course, if only for keeping the research project within manageable proportions, it is necessary to restrict one's attention to one of these questions or a subset of them. In the present paper, I will be concerned only with the representational and processing issues for familiar polymorphemic words in the context of visual word recognition (i.e. reading). Actually, the majority of research papers in the literature have addressed this set of questions. Relatively few researchers have attempted to approach the representation and processing issues in the domain of auditory word recognition (but see Schriefers, Zwitserlood, & Roelofs, 1991; Taft, Hambly, & Kinoshita, 1986; Tyler, Marslen-Wilson, Rentoul, & Hanney, 1988) or in the context of word production (but see MacKay, 1976). Even less attention has been paid to the processing of novel polymorphemic words (but see Coolen, van Jaarsveld, & Schreuder, 1991).
The particular question that will be tackled here is how and why the morphological structure of words could be involved in the kinds of representations and processes used in reading. It seems hard to believe that representations and processes making reference to morpheme-shaped units develop gratuitously, without any motivation. At any rate, I will assume here that if morphology is used as a design principle of the mental lexicon, it must be possible to make sense of the particular design that emanates from it. I will approach the issue by presenting a conceptual analysis of various possible motivations of a morphologically organised mental lexicon, each time discussing the merits and limitations of the account.
REPRESENTATIONAL ECONOMY
The desire to put morphology to the purpose of economic word storage is strongly inspired by the traditional definition of a morpheme. In textbooks on morphology and in introductory courses to general linguistics, morphemes are almost invariably defined as the smallest linguistic units of form and meaning. In accordance with this definition, an English word like boys is said to consist of the stem morpheme boy and the plural suffix -s, as these segments reveal a constant form-meaning relationship across the English lexicon. Examples attesting to their morphemehood are boyish, boyhood, boy's school, boy's book for boy and cats, mugs, dolls for -s.
Can psycholinguists modelling the mental lexicon rely on this traditional definition of the morpheme for making proposals on the way polymorphemic words are represented? It is very tempting to answer this question in the affirmative. Indeed, the definition almost naturally suggests that morphology could be put to the purpose of reducing the amount of information to be stored for these words in the mental lexicon. Two variants of this idea are discussed below with respect to different types of polymorphemic words, and then the explanatory nature of the concept of representational economy is evaluated.
The Mental Lexicon as a Morpheme Inventory
If the above definition is correct, the morpheme rather than the word could be considered as the basic unit of lexical description. If words can be broken down into their constituent morphemes and the latter behave as constants of form and meaning, it must be possible (even desirable) to treat words as simple integrations of morphemic units, at the levels of both form and meaning. Within such a âbuilding-blockâ perspective, the word itself is epiphenomenal and uninteresting for a scientific study of the lexicon. Its formal and semantic aspects being rule-governed, it would seem to claim the same status in the lexical realm that sentences have in syntax. Although such an approach to morphology is suspect in a number of respects (see below), we will further pursue this line of thinking to find out how much sense can be made of it in distinct areas of morphology.
Setting out from the traditional morpheme definition it would take only a small step for the psycholinguist to translate it into a model on the representation and processing of polymorphemic words. Even though written words, when preceded and followed by a space, must be salient units in our visual perception of a piece of text, the definition suggests that morphemes are the salient units in our lexical processing and makes it plausible that polymorphemic words are not stored at all in the mental lexicon. Just like knowledge of the rules enabling language users to generate and understand an infinite number of sentences (Chomsky, 1965), knowledge of the grammar of morphemes might make the presence of lexical representations for polymorphemic words superfluous.
The decision to leave polymorphemic words out of the mental lexicon is made at the representational level but would have important consequences for processing. As a matter of fact, since there would be neither a representation of the words' orthographic form, through which lexical access could be achieved (access representation), nor a representation of their semantic and syntactic properties (linguistic representation), the representational claim would entail a process operating at the level of word form and another process operating at the level of linguistic function. The form-directed process would have to dissect polymorphemic words in order to isolate the morphemes (decomposition process), each of which could then make contact with its access representation. The linguistically oriented process would have to integrate the linguistic representations of the morphemes into a syntactic and semantic representation of the whole word (composition process). Processing would be complete when this fusion had successfully been performed.
The question at hand is whether the concept of a morphemic lexicon with its concomitant processing mechanisms can be applied to the vast area of morphology, only to particular sub-areas, or not at all. Any answer to this question will have to take into account that the term âpolymorphemic wordâ is a common denominator for a class of (more or less) distinct word types, each with its own linguistic properties. We will adopt the traditional tripartite division into inflected, derived and compound words (Matthews, 1974), although linguistically speaking inflectional and derivational morphology might be âfuzzy categoriesâ, regions on a continuum rather than dichotomous sets (see Bybee, 1985). Let us consider the viability of a morphemic lexicon for each of these types separately (derivations and compounds will be treated under the heading of lexical morphology).
Inflectional Morphology
Depending on the richness of the morphology of a language, inflected word forms belong to one or more of the following categories: verbs, determiners, adjectives, nouns, pronouns. Their basic function is a syntactic one, conveying how a word (or the phrase to which it belongs) grammatically relates to other sentence parts (e.g. the German accusative form of the determiner in Er sah den Mann [he saw the man] signals that the noun phrase relates to the verb as direct object) or conveying grammatical agreement with other words (e.g. person agreement as in he plays, gender agreement as in der Mann [the man]).
As inflected forms are syntactically constrained, it is understandable that major portions of the inflectional system of many languages are rule-governed, resulting in the predictability of regular inflected forms. Inflectional affixes are not tied to single lexical items but are part of a morphological paradigm, which is linked to a particular form class (e.g. verbs) and whose slots are defined in terms of syntactic properties (e.g. the inflectional suffix in plays fits within the present-tense paradigm and requires a thirdperson singular subject). Given these properties, it seems apt to treat regularly inflected forms outside the domain of lexical business, as different realisations of a single word (or lexeme). Thus considered, the hypothesis of a morphemic lexicon is highly plausible from the point of view of regularly inflected forms (Chomsky, 1965).
The Impact of the Spelling System. Let us see whether the representational claim derived from this linguistic characterisation of inflectional morphology makes sense as far as its implications for the on-line processing of written words are concerned. At the level of the composition process, no problems would arise. The meaning of affixes in forms like boys and plays can be described as a simple semantic predicate, taking the stem meaning as its argument (PLURAL[BOY], PRESENT[PLAY]), and which would be interpreted by the semantic module in the language system (see Forster, 1979, for a structure of the language processor). The syntactic features of affixes, too, could easily be linked to the syntactic category of the stem and fed into the syntactic module.
Given the definition of the morpheme as a unit with an inalienable form and meaning, one would expect that the decomposition process should also run smoothly and guarantee a straightforward mapping of the orthographic form onto the appropriate morphemic access representations. However, one should take care not to confuse between levels of linguistic description. In linguistics, a distinction is made between an underlying, abstract level and a surface level. The morpheme is a theoretical construct at the abstract level. Although at that level it preserves its (phonological) integrity throughout the set of words in which it appears, this by no means guarantees that it will always surface as the same form. Often, phonological interactions across morphemic boundaries affect the underlying form of a morpheme, yielding âdistortedâ phonetic surface realisations, i.e. allomorphs (e.g. although there is only one plural suffix -s there are three different allomorphs: [S] as in cats, [Z] as in dogs, and [IZ] as in kisses).
Of course, allomorphy in speech does not necessarily correspond to allomorphy in print. The extent to which the spelling system of an (alphabetic) language encodes the surface-level phonetic or the deep-level phonological (and hence morphological) structure of words will determine the degree of allomorphy in the orthography. Unless a spelling system completely abstracts away from the allomorphic variation, a distinction must be made between morphemes and allomorphs (in short, morphs; see Henderson, 1986) when considering the likelihood of morpheme-based lexical access from printed words. The question is whether the reader, who is confronted only with morphs, can travel the route leading from morphs to morphemes and thus arrive at the access representations of the morphemes in regularly inflected forms.
Interestingly, a number of spelling systems seem to have been designed with the purpose of keeping allomorphy at a minimum. English has such a morphographic spelling (Chomsky & Halle, 1968), and so has Dutch. For instance, despite phonetic variation between Dutch paard ([pa:rt]-horse) and paarden ([pa:rdân...