lemma

n. In psycholinguistics, a lemma is a level of representation of a word between the semantic (meaning) and phonological (sound) levels of processing. Lemmas are abstract in that they contain no information about the sound of the word but are uniquely syntactically and semantically specified. The term derives from linguistics, where it refers to the standard form of a word chosen to represent all its possible variants.

The idea of a lemma is most used in speech production, in the context of a two-stage model of lexical retrieval. Semantic specifications are used to retrieve lemmas, which in turn are used to access phonological forms (sometimes called lexemes), although the process of phonological retrieval might itself be a multistage one. Specifying the lemma is called lexical selection, and specifying the phonological form is called phonological encoding. There is considerable evidence to support the two-stage model, with data from speech errors (where we find whole word substitutions related to the target by meaning or sound), the existence of semantic and phonological types of anomia (word-finding difficulties in aphasia), picture-naming studies, and brain imaging. When we are in a tip-of-the-tongue state for a word, we have retrieved the lemma but have not been able to retrieve its associated phonological form (although we might be able to retrieve some of it, such as the first sound of the word). The best evidence for the existence of lemmas is from studies that show that people in tip-of- the-tongue states or with anomia can retrieve gender information about nouns (in those languages that mark gender). Needless to say, there is some dissent as to whether lemmas are really necessary, and there are alternative interpretations of the data.

There have been two large but related debates about how we retrieve and use lemmas in speech production. The first is whether lexicalization (lexical retrieval in speech production) is a discrete or cascading process: that is, do we specify just one lemma before we can start retrieving phonological forms, or are a number of candidate lemmas (related semantically) still active simultaneously while we start to retrieve phonological forms? The second issue is whether or not there is feedback between levels: can the sound of a word affect lemma selection, and in turn can this influence which meaning representations are active? The main evidence, from primed picture-naming studies, is equivocal, and modeling in any case shows that cascading interactive models of lexical retrieval can reproduce any pattern of priming data.

Lemmas can be thought of as corresponding to the hidden level of units in a connec- tionist model of lexicalization.

- TH