B. Krenn and C. Samuelsson. The linguist's guide to statistics. University of the Saarland, December 1997. compendium for a course in Statistical Approaches in Computational Linguistics. 23  Home/Search   Document Not in Database   Summary   Related Articles   Check

....[CG96] or backing off [Kat87] have to be used to alleviate this. 3. Lexicalization is not compatible with the current assignment of probabilities since they can be associated only with respect to the syntactic component of lexical entries. The criteria should be For an introduction see [Cha97, KS97] N grams can be viewed as the simplest kind of statistical language model. In their most basic incarnation, they give for each word a probability of its occurrence that is conditioned by n 1 preceding words. Usually, n is either 2 in which case one speaks of a bigram language model, or 3 ....

....symbol (in accordance with the OSP) following a transition to a new state (in congruence with the STP) and finally returning to the next to last step, or stopping. The main use of HMMs is in analysis (like finding the most probable part of speech assignment for a sequence of words) see [Rab89, KS97] for a classical introduction to HMMs from an NLP point of view) response, whereas a total failure might only be accepted for sufficiently distorted input [Men95] The big asset of this definition, which is close to that for graceful degradation, is that it does not assume any details about ....

B. Krenn and C. Samuelsson. The linguist's guide to statistics. University of the Saarland, December 1997. compendium for a course in Statistical Approaches in Computational Linguistics. 23

....approach in spite of discouraging results and to abandon techniques which can be profitably adapted to new perspectives. 2 The transposition of words can be seen as a combination of insertion and delection, i.e. a combination of superfluous and missing words. 3 For an introduction see [Cha97, KS97] 4 N grams can be viewed as the simplest kind of statistical language model. In their most basic incarnation, they give for each word a probability of its occurrence that is conditioned by n Gamma 1 preceding words. Usually, n is either 2 in which case one speaks of a bigram language model, ....

....symbol (in accordance with the OSP) following a transition to a new state (in congruence with the STP) and finally returning to the next to last step, or stopping. The main use of HMMs is in analysis (like finding the most probable part of speech assignment for a sequence of words) see [Rab89, KS97] for a classical introduction to HMMs from an NLP point of view) 11 word level, semantic information has been incorporated into the lexical analyzer (see [DKIZ95] and combining the above theories with them, it is possible to solve syntactic ambiguities during the parsing process. It is worth ....

B. Krenn and C. Samuelsson. The linguist's guide to statistics. University of the Saarland, December 1997. compendium for a course in Statistical Approaches in Computational Linguistics.

....namely, HPSG parser, probabilistic CFG LR parser, chunk parser and a fall back HMM based 3 dialogue act recognizer. When no parser is able to produce an analysis for the whole input, a module called robust semantic processing tries to combine partial analyses 1 For an introduction see [10, 21]. 2 N grams can be viewed as the simplest kind of statistical language model. In their most basic incarnation, they give for each word a probability of its occurrence that is conditioned by n Gamma 1 preceding words. Usually, n is either 2 in which case one speaks of a bigram language model, or ....

....an output symbol (in accordance with the OSP) following a transition to a new state (in congruence with the STP) and nally returning to the next to last step, or stopping. The main use of HMMs is in analysis (like nding the most probable part of speech assignment for a sequence of words) see [28, 21] for a classical introduction to HMMs from an NLP point of view) 3 produced by the four parsers. Even if this cannot be considered a real distributed architecture (the robust semantic processing is more of an intelligent pipelining technique) moving towards a cooperative model of processing ....

B. Krenn and C. Samuelsson. The linguist's guide to statistics. University of the Saarland, December 1997. compendium for a course in Statistical Approaches in Computational Linguistics. 8

....that the best model of a linguistic phenomenon is the phenomenon itself. Thus, analysis becomes a measurement of how much a text fits into that model. It is clear that if an uttered sentence fails to be analyzed by such a parser, it is a model fault which is too 1 For an introduction see [Cha97, KS97] 2 N grams can be viewed as the simplest kind of statistical language model. In their most basic incarnation, they give for each word a probability of its occurrence that is conditioned by n Gamma 1 preceding words. Usually, n is either 2 in which case one speaks of a bigram language model, ....

....symbol (in accordance with the OSP) following a transition to a new state (in congruence with the STP) and finally returning to the next to last step, or stopping. The main use of HMMs is in analysis (like finding the most probable part of speech assignment for a sequence of words) see [Rab89, KS97] for a classical introduction to HMMs from an NLP point of view) 3 Morphology Robustness at this level is required in order to repair ill formed utterances cause by the following reasons: ffl Speech recognition errors: These errors are generated by the speech recognition module that in certain ....

B. Krenn and C. Samuelsson. The linguist's guide to statistics. University of the Saarland, December 1997. compendium for a course in Statistical Approaches in Computational Linguistics.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC