In psycholinguistics, a lemma (plural lemmas or lemmata) is an abstract conceptual form of a word that has been mentally selected for utterance in the early stages of speech production. A lemma represents a specific meaning but does not have any specific sounds are attached to it.
When we produce a word, we are essentially turning our thoughts into sounds, a process known as lexicalisation. In many psycholinguistic models this is considered to be at least a two-stage process. The first stage deals with semantics and syntactics; the result of the first stage is an abstract notion of a word that represents a meaning and contains information about how the word can be used in a sentence. It does not, however, contain information about how the word is pronounced. The second stage deals with the phonology of the word; it attaches information about the sounds that will have to be uttered. The result of the first stage is the lemma in this model; the result of the second stage is referred to as the lexeme.
This two-staged model is the most widely supported theory of speech production in psycholinguistics[1], although it has been recently challenged.[2] For example, there is some evidence to indicate that the grammatical gender of a noun is retrieved from the word's phonological form (the lexeme) rather than from the lemma.[3] This is easily explained by Caramazza's Independent Network model, which does not assume a distinct level between the semantic and the phonological stages (so there is no lemma representation); in this model, syntactic information about the word in this model is activated in the semantic or phonological level (so gender would be activated in the latter).[4]
The concept of lemma is similar to the Sanskrit sphota (6th c.), an invariant mental word, of which the sound is a feature.