This is not an easy task to quickly design a semantic parser dedicated to a particular task. Indeed, analysis components and linguistic resources are often defined with mutually incompatible formats, which make their assembly complex. We wish to bring an operational response to this problem with the Antelope linguistic platform, whose design and implementation principles are described in this thesis. Inspired by the Meaning-Text Theory (MTT), Antelope targets a robust syntactic and semantic parsing of texts, and can handle large corpora; its goal is to enable deep understanding of various kinds of text: consumer reviews, articles from encyclopedia, HR documents, newspaper articles... To achieve this goal, Antelope integrates (i) several ready-to-use components, addressing the most common NLP tasks, which interact within a unified text analysis model; (ii) a broad-coverage multilingual semantic lexicon compiled from various sources. An integration effort of all these components provides a robust and homogeneous platform, with a syntax-semantics interface. The thesis presents the platform and compares it with other state-of-the-art projects; it highlights the best practices that should be taken to ensure that such complex software remains maintainable; it also introduces a semi-supervised approach for large-scale knowledge acquisition.
