A comparative study between polyclass and multiclass language models

In this work, we introduce the concept of Multiclass for language modeling and we compare it to the Polyclass model. The originality of the Multiclass is its capability to parse a string of classes/tags into variable length independent sequences. A few experimental tests were carried out on a class corpus extracted from the French « Le Monde » word corpus labeled automatically. This corpus contains a set of 43 million of words. In our experiments, Multiclass outperform first-order Polyclass but are slightly outperformed by second-order Polyclass.