Combination of Language Models for Word Prediction: An Exponential Approach

• Published in: IEEE/ACM transactions on audio, speech, and language processing Vol. 24; no. 9; pp. 1481 - 1494
• Main Authors: Cavalieri, Daniel C., Palazuelos-Cagigas, Sira E., Bastos-Filho, Teodiano F., Sarcinelli-Filho, Mario
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.09.2016; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Combination of language models; Initial conditions; Interpolation; Mathematical analysis; Mathematical model; Mathematical models; Maximum entropy; Natural language processing; natural language processing (NLP); Predictive models; Speech; Speech processing; Tasks; word prediction (WP); natural language processing (NLP); word prediction (WP); Combination of language models
• ISSN: 2329-9290; 2329-9304
• DOI: 10.1109/TASLP.2016.2547743

This paper proposes an exponential interpolation to merge a part-of-speech-based language model and a word-based <inline-formula><tex-math notation="LaTeX">n</tex-math></inline-formula>-gram language model to accomplish word prediction tasks. In order to find a set of mathematical equations to properly describe the language modeling, a model based on partial differential equations is proposed. With the appropriate initial conditions, it was found an interpolation model similar to the traditional maximum entropy language model. Improvements in keystroke saved and perplexity over the word-based <inline-formula><tex-math notation="LaTeX">n</tex-math></inline-formula>-gram language model and two other traditional interpolation models is obtained, considering three different languages. The proposed interpolation model also provides additional improvement in hit rate parameter.