Kuhn, R. and R. de Mori. 1990. A cache-based natural language model for speech recognition. IEEE Trans. on Pattern Analysis and Machine Intelligence, 12:570--583.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and philosophy. Empirical systems perhaps offer a ray of hope in this regard, since they are based on learning, and learning (about new contexts) is seen to be a key element missing from the way language is conceived in the traditional symbolic approach. Work on adaptive statistical modeling, eg [73, 106], may constitute a rudimentary step in this direction. But the potential of empirical approaches should not be overstated, because their successes so far have been mostly practical, and there is no hard evidence that they can solve the dicult problems which have defeated rationalist methods. To ....

Roland Kuhn and Renato De Mori. A cache-based natural language model for speech recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 12(6):570-583, June 1990.

....with 134 tags used in this work and described below, and its successor CLAWS2a with 179 tags. Stochastic processing has typically made use of the much larger tagsets for corpus based analysis. De Marcken uses the 132 LOB corpus tagset for his simple parser [37, 1990] The work of Kuhn and de Mori [38, 1990] on the linguistic probability of word strings uses the 153 tags defined by Johansson [39, 1989] The IBM Lancaster Research Group use the 179 CLAWS2a tagset for their automated parsing project, but the designers are currently looking at reducing the size of the tagset to about 60 items which are ....

R Kuhn and R de Mori. A cache based natural language model for speech recognition. IEEE trans, on Pattern Analysis and Machine Intelligence, 1990.

....Block occurrences in LZW dictionary higher than that of the intermediate blocks. This can be regarded as a typical English text file characteristic and is in conformance with the evidence that if a word occurs once in an English text, then there is a high probability that it will occur again soon [3]. 3.2 Dynamic Block Shifting If the distribution of codewords for E.coli can be altered in such a way that the block of codewords occurring most frequently is shifted to the end of the dictionary, small improvement can be expected when using the PB or DS codes. This simple procedure, termed as ....

R. DeMori and R. Kuhn. A cache-based natural language model for speech recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(6):570--583.

....Block occurrences in LZW dictionary higher than that of the intermediate blocks. This can be regarded as a typical English text le characteristic and is in conformance with the evidence that if a word occurs once in an English text, then there is a high probability that it will occur again soon [3]. 3.2 Dynamic Block Shifting If the distribution of codewords for E.coli can be altered in such a way that the block of codewords occurring most frequently is shifted to the end of the dictionary, small improvement can be expected when using the PB or DS codes. This simple procedure, termed as ....

R. DeMori and R. Kuhn. A cache-based natural language model for speech recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(6):570-583.

....corpora consisting of many different topics. On the test corpus, they do not adapt their probabilities according to the topic of the actual test data. Most often this problem is addressed by a cache language model which is made an additional component of interpolated language models, as in [4] and [8]. Additionally, in this paper we interpolate topic dependent language models with the baseline model and perform an adaptation by putting more weight on that topic dependent language model which is most promising given the current history, as in [7] Further recent approaches to adaptive ....

R. Kuhn, R. de Mori: "A Cache--Based Natural Language Model for Speech Recognition", IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol. 12, pp. 570--583, June 1990.

....are based on a training corpus which may not match the speaker s more limited vocabulary. The trigram model is unable to adapt to the speaker s vocabulary and style, nor to topic information which is contained in a document. An improvement upon trigram model was proposed in Kuhn and De Mori [15] and Jelinek et al. 13] The cache model takes into account the relative dynamic frequencies of words in text seen so far and incorporates this information into the prediction by maintaining a cache of recently used words. This results in up to a 23 improvement in perplexity as reported in ....

....5. Performance of Smoothed ME N grams Models Model Size of Model Perplexity Deleted Interpolation 3.5M 225.1 ME w threshold of 2 790K 247.9 Fuzzy ME w threshold of 2 790K 230.3 Good Turing Discounted ME w threshold of 2 790K 220.9 6. 0 Self Triggers The success of the cache in Kuhn and De Mori [15] suggests that the appearance of a word has a substantial impact on the probability of a future appearance of the same word. Preliminary experimentation in Lau [17] and Lau et al. 18] reveal the same interesting phenomenon. We will refer to the effect of a word on its own future probability as ....

[Article contains additional citation context not shown here]

Kuhn, R. and De Mori, R., "A Cache-Based Natural Language Model for Speech Recognition," in IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(6): 570-583, 1990.

....however, all experiments in this paper use trigram models. Although quite powerful given their simplicity, n gram models are constrained in their inability to take advantage of dependencies longer than n. One approach to overcome this limitation is to use dynamic cache language models [1, 2, 3], which model tasklevel dependencies by increasing the likelihood of a word given that it has been observed previously. However, cache models do not account for dependencies within a sentence. Context free grammars [4, 5] could account for grammatical dependencies within a sentence; however, it is ....

....can be extended in several ways. One approach to overcome fragmentation problems in very sparse data domains could include using an n gram part of speech sequence model as the base for all component models and topic dependent word likelihoods given the part of speech labels, a natural extension of [2], assuming that the part of speech sequence probabilities are not topicdependent and can be based on the entire training data. The simple static mixture language model can also be useful in applications other than continuous speech transcription. For example, topicdependent models could be used ....

R. Kuhn and R. de Mori, "A Cache Based Natural Language Model for Speech Recognition", IEEE Transactions PAMI, Vo 14, pp. 570-583, 1992.

....on the compression of a PPM model would be very interesting. Despite the foregoing, there are classes of information that are not taken into account by PPM (or by other related techniques) There is evidence, for example, that in English text there are words which show strong recency effects [24]. If the word occurs once then there is a high probability that it will occur again soon. It is unclear, at this time, what models and estimation techniques can effectively take advantage of such information. ....

DeMori, R. and Kuhn, R. (1990) A cache-based natural language model for speech recognition. IEEE Trans. Patt. Anal. Machine Intell., PAMI-12, 570--583.

....classes. Approaches that change language models according to the estimated topic class M (e.g. 9] correspond to this formulation. Approaches using probabilistic state transition networks [10] or HMM [11] for forming semantic language models are also classified into this category. A cache model [12] is one of the approaches in which M is implicitly represented. In this paper, we consider that M is represented by a co occurrence of words based on the distributional hypothesis by Harris [13] Similar methods include a method that uses a thesaurus for measuring semantic similarity between ....

R. Kuhn and R. De Mori, "A Cache-based Natural Language Model for Speech Recognition," IEEE Trans. PAMI-12, 6, pp. 570-583, 1990.

....occur in C i.e. their frequency of occurrence in C is 0. Researchers in speech and language processing have tended to avoid letting this theoretical limitation restrict them and have concentrated their efforts on casting the problem as a technical one and attempting to find solutions to it [89, 95, 85, 31]. One of the most important structural features of a corpus of natural language concerns the relationship between the frequency of occurrence of a particular word segment and the length of that segment. In a corpus of N words, there are N sets of segments, each set containing segments of the same ....

....model represents one often used and powerful approach to generating equivalent contexts the recent word context equivalence class. It is commonly used partly because it involves an easy method of generating useful equivalent contexts whilst at the same time improving predictive power [95]. The use of equivalence classes allows a more general theoretical description of contexts. For example, the trigram models, which provided such a spur to research [32] can now be re described in terms of that set of equivalence classes which unites all contexts which share the same last and ....

[Article contains additional citation context not shown here]

Ronald Kuhn and Renato De Mori. A cache-based natural language model for speech recognition. I.E.E.E. Transactions on Pattern Analysis and Machine Intelligence, 12(6):570 -- 583, June 1990.

....is the cache based model presented in the following section. 3.2. Cache based model This model consists in storing, for each singular plural homophone word, its left contexts as seen in the training corpus. These contexts are word histories made of the last ten words stored in a cache memory [6]. Each cache content C(w) is a vector whose components are the syntactic POSs assigned to the words by the tagger. The size of the vectors corresponds to the number of POS which is 105. The training of this model consists in using the training corpus for updating two cache memory vectors for each ....

Kuhn R., De Mori R (1990) A Cache Based Natural Language Model for speech Recognition IEEE Trans. Pattern anal. Machine Intell., PAMI12 (6):570-582.

....than a for statement. Once the system is used and results are evaluated these probabilities can be adjusted to improve the performance. Probabilities can be dynamically adapted to a specific software system using a cache memory method originally proposed (for a different application) in [Kuhn90]. A cache is used to maintain the counts for most frequently recurring statement patterns in the code being examined. Static probabilities can be weighted with dynamically estimated ones as follows: P (S i jA j ) Delta P cache (S i jA j ) 1 Gamma ) Delta P static (S i jA j ) In this ....

Kuhn, R., DeMori, R., "A Cache-Based Natural Language Model for Speech Recognition ", IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 12, No.6, June 1990, pp. 570-583.

No context found.

Kuhn, R. and R. de Mori. 1990. A cache-based natural language model for speech recognition. IEEE Trans. on Pattern Analysis and Machine Intelligence, 12:570--583.

No context found.

R. Kuhn and R. de Mori. A cache-based natural language model for speech recognition. IEEE Trans. on Pattern Analysis and Machine Intelligence, 12:570--583, 1990.

No context found.

Roland Kuhn and Renato De Mori. A Cache-Based Natural Language Model for Speech Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, volume PAMI-12, number 6, pages 570--583, June 1990.

No context found.

Roland Kuhn and Renato de Mori. A Cache-Based Natural Language Model for Speech Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(6):570--583, 1990.

No context found.

Roland Kuhn and Renato de Mori. A Cache-Based Natural Language Model for Speech Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(6):570--583, 1990.

No context found.

R. Kuhn and R. De Mori, "A cache-based natural language model for speech recognition," IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 12, no. 6, pp. 570--583, June 1990.

No context found.

R. Kuhn and R. De Mori. A cache-based natural language model for speech recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(6):570--583, June 1990.

No context found.

R. Kuhn and R. De Mori. A cache-based natural language model for speech recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(6):570--583, June 1990.

No context found.

R. Kuhn and R. de Mori, "A Cache-Based Natural Language Model for Speech Recognition", IEEE Trans. PAMI, Vol. 12, pp. 570-583, June 1990.

No context found.

R. Kuhn and R. de Mori, "A Cache Based Natural Language Model for Speech Recognition", IEEE Transactions PAMI, Vol. 14, pp. 570-583, 1992.

No context found.

R. Kuhn and R. de Mori. "A Cache Based Natural Language Model for Speech Recognition. " IEEE Transactions on Pattern Analysis and Machine Intelligence, volume 14, pages 570--583, 1992.

No context found.

R. Kuhn and R. D. Mori. A cache-based natural language model for speech recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, PAMI-12(6):570--583, 1990.

No context found.

R. Kuhn and R. D. Mori. A cache-based natural language model for speech recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, PAMI-12(6):570--583, 1990.

First 50 documents  Next 50

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