Introduction

How humans process language quickly and efficiently remains largely unexplained. The main difficulty is that, although many disciplines (linguistics, psychology, computer science, and neuroscience) have addressed this question, it is difficult to describe language as a global system. Typically, no linguistic theory entirely explains how the different sources of linguistic information interact. Most theories, and then most descriptions, only capture partial phenomena, without providing a general framework bringing together prosody, pragmatics, syntax, semantics, etc. For this reason, many linguistic theories still consider language organization as modular: linguistic domains are studied and processed separately, their interaction is implemented at a later stage. As a consequence, the lack of a general theory of language, accounting for its different aspects, renders difficult the elaboration of a global processing architecture. This problem has direct consequences for natural language processing: the classical architecture relies on different subtasks: segmenting, labeling, identifying the structures, interpreting, etc. This organization more or less strictly imposes a sequential view of language processing, considering in particular words as being the core of the system. Such a view does not account for the fact that language is based on complex objects, made of different and hetergeneous sources of information, interconnected at different levels, and which interpretation cannot always be done compositionally (each information domain transferring a subset of information to another).