Abbreviation Recognition with MaxEnt Model

Abbreviated words carry critical information in the literature of many special domains. This paper reports our research in recognizing dotted abbreviations with MaxEnt model. The key points in our work include: (1) allowing the model to optimize with as many features as possible to capture the text characteristics of context words, and (2) utilizing simple lexical information such as sentence-initial words and candidate word length for performance enhancement. Experimental results show that this approach achieves impressive performance on the WSJ corpus.

• Type: Conference
• Year: 2006
• Where: CICLING
• Authors: Chunyu Kit, Xiaoyue Liu, Jonathan J. Webster
• Keywords: Abbreviated Words, Candidate Word Length, CICLING 2006, MaxEnt Model, Natural Language Processing