Nominalizations, i.e. nouns morphologically derived from verbs (e.g. construction) or from adjectives (e.g. sadness), have continued to occupy a central place in grammatical analysis, but linguists have limited their investigation almost exclusively to relationships involving the participant roles in the described situations and the linguistic representation of those roles (Grimshaw 1990). Very little attention has been paid to the fundamental question of the extent to which the temporal and aspectual information that a verb or adjective carries is present in the semantics of a related noun, or to how that information should be captured in a semantics with wide empirical coverage. The proposed project will contribute to redressing this situation, both on an empirical and on a theoretical level. On the empirical level, we aim at capturing occurrences of such nominalizations from reference corpora, first in French, and then in other languages, using such reference corpora as the French Treebank (Abeillé, 2003) and the Negra Corpus for German. The first step in our project is thus to collect a reasonable amount of naturally occurring examples of deverbal and deadjectival nouns, as opposed to artificially constructed ones, in order to broaden the scope of the proposed semantic analysis. For this task, we will rely on both symbolic (rule-based) and machine-learning strategies, building on the accumulated observations and knowledge rendered possible by past working sessions of PAI Ontoref. This data collection phase is crucial to the subsequent steps of our project, and will ensure that the final goal, a semantics-driven nominalization lexicon, will be achievable. On the theoretical level, the most relevant question we want to answer is to what extent nominalizations inherit semantic (in particular, aspectual) features from the verbs or adjectives from which they derive. In this respect, the main problem we are faced with is finding appropriate linguistic tests to determine this, as most of those used for verbal predicates (Vendler, 1967; Dowty, 1979) do not apply to nouns. In previous research (Huyghe & Marín, 2006), we have proposed several tests that appear to work adequately. From the preliminary results already obtained, we can assert that while some aspectual features are indeed inherited by the nominal form, there are certain mismatches. Detecting such discrepancies, together with finding a reliable set of aspectual diagnostics, are two of the main goals of the proposed project. In addition to improving our understanding of the relation between the Aktionsart of nouns and morphologically related verbs and adjectives, we also plan to determine the kind of natural language ontology that must be posited for reference to the sorts of abstract objects (situations, facts, propositions) that nominalizations denote (Asher, 1993; Zuchi, 1993; Ginzburg & Sag, 2001). Researchers mostly agree that a better understanding of nominalizations is crucial for NLP tasks such as ontology development or information retrieval, but again approaches based on argument structure (Meyers et al., 1998) have clearly predominated. In contrast, we propose to develop a semantics-driven nominalization lexicon, integrating as much explicit knowledge on argument structure as possible, but supplementing it with as much semantic information as possible, collected and formalized through the different preceding stages of the project. Another expected outcome of the project is the testing and distribution of a semantic annotation tool. This tool will use the accumulated and formalized knowledge on nominalizations as represented in our proposed lexicon, as well as XCRF, a general tree-annotation tool distributed by GRAPPA-LIFL (Jousse et al., 2006). This will make it possible to not only semantically tag occurrences described in our lexicon, but also unknown occurrences, thus achieving wide empirical coverage.