Theoretically, the assumption of same knowledge sources for both generation and understanding is essential for the common view of language as an interpersonal medium and an interface to thought [McDonald1987], i.e. if communication is to take place, a correspondence must be established between the mental representation of an utterance for a speaker and that for a listener. Consequently, any model of language behaviour that hypothesises only one linguistic representation is preferable to one where two different representations are used that are only applicable in one of the major modes (either understanding or generation). The fact that humans can understand meaningful utterances they have produced intentionally indicates that a subset of the internal representations used during understanding and generation must be the same. Intuitively, this must be the case at least for the mental representation of an utterance; otherwise it would be very hard to explain why humans are able to paraphrase what they have said or heard.
But there are also arguments that motivate the assumption that linguistic entities, i.e. grammatical structures are shared during understanding and generation. Garrett Garrett:82 argues that if it would be possible to discover that understanding and generation can be modelled in substantially the same way, one might interpret this as an evidence for separating the informational structure of a language (e.g., the rule system and the lexical component of a grammar) from modality-specific aspects of the language. This declarative aspect of grammatical knowledge of a language is very important for illuminating the relation between grammatical theory and processing theory. For example, Fodor and Frazier FodorFrazier:80, argue at length that non-declarative grammars (like the ATN framework, cf. [Woods1986]) are inappropriate for explaining the interaction of different parsing strategies, specifically when the parser's preference for low attachment and its preference for minimal attachment are in conflict. Frazier concludes that ``the claim of shared syntactic knowledge is at least coherent and consistent with available evidence concerning the mental representation of syntactic knowledge `` (Frazier Frazier:82, page 229). There is not only empirical evidence for shared syntactic knowledge but also that during understanding and generation lexical information are shared and that the actual routines used in lexical retrieval are common in both processes (see [Garrett1982]).