🧑‍🔬 Ted Pedersen

Ted Pedersen works on lexical semantics and on the evaluation of language technology. The papers here ask how closely two concepts are related in meaning and how systems that link words across languages can be compared fairly. The work draws on lexical resources such as WordNet and on parallel texts, and it pairs new measures with structured comparative testing.

The most cited paper here, Extended Gloss Overlaps as a Measure of Semantic Relatedness, written with Satanjeev Banerjee, builds on the Lesk idea of counting words that dictionary definitions share. WordNet glosses are short, so the measure also compares the definitions of neighbouring concepts, such as hypernyms, hyponyms and meronyms, along with usage examples. It gives longer shared phrases more weight than scattered single words. The second paper, An Evaluation Exercise for Word Alignment, with Rada Mihalcea, reports on a shared task held at a 2003 HLT/NAACL workshop on parallel texts. Teams aligned every token of a test set across sentence-aligned corpora in two language pairs. The paper describes the setup, the participating systems and their results. Its design kept systems that used outside resources apart from those that did not, and it scored alignments both with and without unlinked words.

Pedersen's papers here date from 2003 and touch on Machine translation and Evaluation & benchmarks. Pedersen is based at the University of Minnesota, Duluth. Earlier bylines also list the University of Minnesota and, in the late 1990s, Southern Methodist University.

Ted Pedersen has 2 papers in our library, in 2003, most of them on Machine translation, Evaluation & benchmarks; often with Rada Mihalcea, Satanjeev Banerjee.