Models of Second Language Production

The article discusses to what extent bilingual models of speech production can account for the fact that L2 speakers are generally, but not always, able to select LI words when speaking the LI and L2 words when speaking the L2. After a brief discussion of word selection in monolingual models of speech production, two bilingual models are presented. In one of these, LI and L2 systems form separate subsets which are activated in their entirety. In the other one there is only one network containing both LI and L2 items, and the language feature plays a role in activating individual lexical items. It is argued that the second representation has the advantage of being more efficient, because it allows one to explain the selection of a single LI word to be used in L2 speech without having to activate the entire LI system. The first model cannot give an efficient explanation for the language switches violating the matching principle proposed by Bierwisch & Schreuder (1993), which entails that a lemma is selected only if it includes all and only those of the semantic features contained in the chunk to be lexicalized. The article also addresses the chunking problem. Since there is no one-to-one correspondence between concepts and words, speakers need to cut up their messages into lexicalizable chunks. It is suggested that both LI and L2 speakers may follow an informed trial-and-error procedure based on their experiences with language(s) involved. Since there is no look-ahead, speakers do occasionally rum into lexical problems. In view of the matching principle, the only way to solve these problems is to go through the processes of conceptualization and chunking again.