Collecting Statistical Information on Noun-Adjective Multiword Expressions for Extracting the Noun-Noun Ones

Extraction and validation of multiword expressions (MWEs) thanks to association measures is a very common method. Researchers use it for extraction of MWEs of different lengths or different syntactic structures. However, we never wonder if the association measures collected on an MWE with one particular structure (e.g. Verb-Preposition) is relevant for collecting an MWE with a different one (e.g. Adjective-Noun). In this article we use Noun-Adjective MWEs association measures for training a model that is evaluated on the task of Noun-Noun MWEs extraction and validation. We work on the French part of the Europarl corpus and use the dictionary Dela as our gold standard. At the end we show that trained on the same text the model tuned with Noun-Adjective candidates is better than the model tuned on Noun-Noun candidates for extracting Noun-Noun MWEs.