7.1　Mental representation of bilingual lexicon

Lexical knowledge in language production

The implementation of lexical knowledge in the mind/brain of the individual speaker/hearer constitutes what is usually called the mental lexicon（ML）.Besides variation corresponding to the differences in individual knowledge.ML must be assumed to contain additional information，reflecting，for example，frequency of use or familiarity.Information in the mental lexicon must furthermore be organized according to the conditions of use in comprehension and production.

Instead of speculating about more appropriate models of the mental lexicon，we will merely indicate by means of the following schema（see Figure 7.9）the interaction of different components，which any model must account for in one way or the other.This schema，which elaborates on Levelt's general model of language production，must not be considered as an actual flow chart，as it only represents the way in which different types of information and mapping processes depend on each other.

Figure 7.9　The interaction of different components in language production.Source：Willem（1991）

This schema is simplified in several respects，in order to keep it perspicuous for the present discussion.Thus，the way in which morphological aspects of actual words relate to the lexicon and enter the formulator was left out.Some of the relevant categories，such as tense and number，have a clear and extensively studied conceptual interpretation and must therefore originate from CS.Other morphological categories，such as agreement or structural case are a matter of purely grammatical，language internal structure and bear no direct relation to CS.The fact that both verbalization and interpretation are sensitive to the distinction between the context of（a part of）an utterance，and its conceptual purport in the narrower sense was not indicated.

One has nothing to say about the various sorts of input to the conceptualizer and the way in which it generates a conceptual structure CS.In the light of what has been said earlier，one has to assume，however，that there is a mechanism that takes CSasinput and realizes the mapping Vbl.This mechanism has to affect two things：first，it must split up CS into chunks that can be lexicalized.thereby registering the relation between them and making them available for the formulator；second，it must feed chunks into the system of lemmas where they must be matched with the semantic form SF（Ei）of appropriate lexical entries Ei.The selection of some SF（Ei）is directly connected with the determination of the argument structure AS（E1）and the grammatical features GF（Ei）.Both of which must be available to the formulator，which constructs the syntactic structure SS in accordance with the relational information provided by Vbl.The integrated semantic form SF，compiled from the SF（Ei）of the accessed lexical items of an utterance，is best understood as a criterion against which monitoring of the result of lexical access and integration into a connected utterance structure is possible.This monitoring must be effected by the interpretation mechanism Int.which also maps SF into CSin language comprehension.

While it is assumed that SF（E1），AS（E1），and GF（E1）are different，but directly related types of information constituting the lemma of E1，the phonetic form PF（E1）—and the graphemic form，for that matter—belongs to the system of lexemes，which is different component of the mental lexicon delivering its information to the articulator.We will not make any further comments on the access of lexemes and their integration.

Two final points should be noted with respect to the schema given above.First，while the formulator and the articulator must be central processing units operating sequentially on the input provided by Vbl to produce SF and SS（and ultimately the articulatory output），this does not hold for the mechanism realizing Vbl.This becomes quite obvious，if we recognize that the lemma component containing SF（Ei），AS（Ei），and GF（Ei）abbreviates the whole system of lemmas，that is，the proper memory basis of the mental lexicon.Second，while assembling an integrated SF and constructing the corresponding SS by the formulator is largely a matter of sequential processing，the matching of conceptual chunks against the lemmas to be accessed is a distributed process that allows for extensive parallel processing.If this assumption is correct，then Vbl does not only sort out conceptual chunks，thereby registering their relationship，but also feeds on the one hand into the sequentially operating formulator，and triggers on the other hand parallel processes on the system of lemmas.