🧑‍🔬 Marc Moens

Marc Moens works in natural language processing, with papers here on summarization and named entity recognition. Both lines of work look at how a system can find the parts of a text that matter. One asks which sentences of a long article carry its contribution. The other asks which strings name people, places and organisations. The methods combine annotated corpora, hand-built rules and statistical models, and the experiments test which resources a task really needs.

The most cited paper here is Summarizing Scientific Articles: Experiments with Relevance and Rhetorical Status, written with Simone Teufel. It argues that plain sentence extraction fails on research papers, because a lifted sentence may describe the authors' own method or a rival one they reject. The paper therefore tags each selected sentence with its rhetorical role. It also builds a human-annotated benchmark from computational linguistics conference papers and checks how far annotators agree. The second, Named Entity Recognition without Gazetteers, uses the MUC-7 newswire data to test how much a recogniser depends on large name lists. It pairs rule-based grammars with maximum entropy models and draws on both the form of a name and its surrounding context. It finds that small lists of well-known names, or lists gathered from a modest amount of in-domain text, serve almost as well as large ones.

The papers here run from 1999 to 2002. They move from extracting entities in news text toward the discourse structure of scientific writing. Both are bylined from the University of Edinburgh, where the author's affiliation spans many years. Co-authors include Simone Teufel and Claire Grover.

Marc Moens has 2 papers in our library, in 1999–2002, most of them on Named entity recognition, Summarization; often with Claire Grover, Simone Teufel.