An adaptive statistical languagemodel is described, which successfullyintegrates long distancelinguistic information with other knowledge sources. Most existing statistical language models exploit only the immediate history of a text. To extract information from further back in the document's history, we propose and use trigger pairs as the basic information bearing elements. This allows the model to adapt its expectations to the topic of discourse. Next, statistical evidence from multiple sources must be combined. Traditionally, linear interpolation and its variants have been used, but these are shown here to be seriously deficient. Instead, we apply the principle of Maximum Entropy (ME). Each information source gives rise to a set of constraints, to be imposed on the combined estimate. The intersection of these constraints is the set of probability functions which are consistent with all the information sources. The function with the highest entropy within that set is the ME solution...

The Zipf curve of log of frequency against log of rank for a large English corpus of 500 million word tokens and 689,000 word types is shown to have the usual slope close to –1 for rank less than 5,000, but then for a higher rank it turns to give a slope close to –2. This is apparently mainly due to foreign words and place names. The Zipf curve for a highly-inflected language (the Indo-European Celtic language, Irish) is also given. Because of the larger number of word types per lemma, it remains flatter than the English curve maintaining a slope of –1 until a turning point of about rank 30,000. A formula which calculates the number of tokens given the number of types is derived in terms of the rank at the turning point, 5,000 for English and 30,000 for Irish.

The Speech Research Group of the University of Granada is actively working in several topics of Speech Science. This paper explain the main guidelines of our work on Automatic Speech Recognition. After a general introduction, we devote several sections to describe our main contributions to this field. 1 Introduction The Speech Research Group from the University of Granada (Spain) is a young Group formed by nine researchers. At the present time, three of them are Doctors and three more are expected to get the Ph. D. degree by the end of this year. The main field of interest of the Group is Speech Recognition and Coding. Six of the members of the Group are concerned with Speech Recognition, while the other three work on Speech Coding. This paper is a short description of the goals and achievements of the Group in the field of Automatic Speech Recognition, which are described in the following Sections. As far as Speech Coding is concerned, we will just mention our work related to: ffl V...

In Statistical Machine Translation, we often find forward or backword jumps while translating from a source position to a target position. We propose several position alignment models for estimating these jump probabilities. Our initial jump probability model is a coarse model with no dependencies. The maximum likelihood estimation of the jump probabilities is performed during post word alignment using maximal approximations from the IBM 4 word alignment model output. We systematically add intutive parameters to provide more accurate models and evaluate these models based on their perplexity on a test set. We also report our results with smoothing via linear interpolation to account for data sparseness issues. The improvement in perplexity can provide an indication of how well the additional parameters model the jump probabilities. The best model can then be applied in various MT components like the decoder or in the iterative word-alignment model training itself.