This paper deals with the construction of lexical semantic resources for predicates, verbs and prepositions. We first raise questions about the theoretical perspectives and the methods to be applied. Next, we describe our resources: alternations, thematic grids and lexical conceptual structure representations. We conclude by some indications on the use of these resources in applications.

How do listeners segment the continuous speech input into words? This paper addresses this important question by reviewing current views of lexical segmentation. After providing some background to the more general problem of spoken word recognition, we examine the different cues - including phonetic, phonological, prosodic, and lexical cues - that have been shown to play a role in speech segmentation. We then analyze how these cues are used to locate alignment points in the continuous speech signal. These points serve to define what parts of the signal are matched with which representations in the mental lexicon. Several solutions to the alignment problem are discussed in light of existing experimental evidence.

The difficulties that arise with collecting neologisms, individually or collectively, manually or automatically, come from the diversity of the terms and the theoretical problems concerning their statute. Excluding a term from a corpus is a delicate question. One risks to leave out a good deal of formal neologisms, and a fortiori, other types of neologisms such as those of a semantic or syntactic character, and those created by lexical combinations or by deviation. Observed variations of what people see as neologisms are not completely random: they are due partly to the people themselves, partly to lexical matrices. One can try to build a scale of neologicity. What is left to be done, is the social acceptation of the neologism, its life span (from when to when?), not considering its unforeseen trajectories.