Pereira, Fernando C. N., Michael Riley, and Richard W. Sproat. 1994. Weighted rational transductions and their application to human language processing. In ARPA Workshop on Human Language Technology. Morgan Kaufmann.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....techniques and generative process techniques should be counterbalanced by the similarities that are brought to the fore when one re expresses rule based taggers as finite state transducers. Namely, HMM s can also be viewed as stochastic finite state transducers, as discussed by Pereira et al. [65]. This line of inquiry promises to give us a model of tagging (and partial parsing, as we shall see) of great generality, and is an area that will likely receive increasing attention. 2 Partial Parsing Let us turn now to parsing. Traditional parsers including standard stochastic parsers aim ....

....state sequence the automaton passes through. In this way it is possible to do stratal parsing with standard HMM training and recognition techniques. In more formal terms, we have turned each stratum automaton into a finitestate transducer, composed the transducers, and eliminated # transitions [48, 71, 65]. The only di#erence from standard transducer composition is that outputs at intermediate levels matter. The standard algorithms assume that states may be merged if doing so does not a#ect the relationship between the input and the final output. But in stratal parsing, we wish to keep states ....

Fernando C.N. Pereira, Michael Riley, and Richard W. Sproat. Weighted rational transductions and their application to human language processing. In Human Language Technology Workshop, pages 262--267, 1994.

....the distribution of an output sequence when given an input sequence of the same length, using a hidden state variable and a Markovian independence assumption, as in Hidden Markov Models (HMMs) Rabiner, 1989] in order to simplify the distribution. IOHMMs are a form of probabilistic transducers [Pereira et al. 1994, Singer, 1996] with input and output variables which can be discrete as well as continuous valued. However, in many sequential problems where one tries to map an input sequence to an output sequence, the length of the input and output sequences may not be equal. Input and output sequences could ....

Pereira, F., Riley, M., and Sproat, R. (1994). Weighted rational transductions and their application to human language processing. In ARPA Natural Language Processing Workshop.

....database of complex terms can be augmented and enriched. 5 Finite State Implementation of the Derivational System Our model is based on generative morphology (Aronoff1976; Selkirk1982) and inspired by the work of (Corbin1987) for French; the implementation uses finite state transducer (FST) tools (Pereira, Riley, and Sproat1994; Mohri and Sproat1996) The system is written to handle French morphology, but the algorithms and compilers are language independent, and have been applied to several other languages (Sproat1996) Like all finite state transducer systems, the morphological system is capable of generation and ....

Pereira, Fernando, Michael Riley, and Richard Sproat. 1994. Weighted rational transductions and their application to human language processing. In ARPA Workshop on Human Language Technology, pages 249--254. Advanced Research Projects Agency, March 8--11.

.... 14] to natural language processing [17, 18, 24, 26] A subclass of finitestate machines, the weighted finite state automata (WFAs) has recently assumed new importance, because WFAs provide a powerful method for manipulating models of human language in automatic speech recognition (ASR) systems [19, 20]. This new research direction also raises a number of challenging algorithmic questions [5] A weighted finite state automaton (WFA) is a nondeterministic finite automaton (NFA) A, that has both an alphabet symbol and a weight, from some set K, on each transition. Let R = K; Phi; Omega ; 0; ....

....out of each state. Not all rational power series can be generated by deterministic WFAs. A determinization algorithm takes as input a WFA and produces a deterministic WFA that generates the same rational power series, if one exists. The importance of determinization to ASR is well established [17, 19, 20]. As far as we know, Mohri [17] presented the first determinization procedure for WFAs, extending the seminal ideas of Choffrut [7, 8] and Weber and Klemm [27] regarding string to string transducers. Mohri gives a determinization procedure with three phases. First, A is converted to an equivalent ....

[Article contains additional citation context not shown here]

F. Pereira, M. Riley, and R. Sproat. Weighted rational transductions and their application to human language processing. In Proc. ARPA HLT, pages 249--54, 1994.

....a mixture of finite state and context free techniques. They use NP rules of a pruned treebank grammar. For processing, each point of a text is matched against the treebank rules and the longest match is chosen. Cascades of automata and transducers can also be found in speech processing, see e.g. (Pereira et al. 1994; Mohri, 1997) Contrary to finite state transducers, Cascaded Markov Models exploit probabilities when processing layers of a syntactic structure. They do not generate longest matches but most probable sequences. Furthermore, a higher layer sees different alternatives and their probabilities for ....

Fernando Pereira, Michael Riley, and Richard Sproat. 1994. Weighted rational transductions and their application to human language processing. In Proceedings of the Workshop on Human Language Technology, San Francisco, CA. Morgan Kaufmann.

....function constitute a process cost model, or simply a model. The costs specified by the cost function are the model parameters; the other components constitute the model structure. We contrast our costed event framework to a related effort toward providing tools for processing weighted automata [13]. The weighted automata tools provide algorithms for manipulating labeled, weighted, state transition networks. We further separate the algorithmic issues from those relating to cost manipulation. For example, the implementation of a speech recognizer using the weighted automata tools would be an ....

F. C. N. Pereira, M. Riley, and R. W. Sproat. Weighted rational transductions and their application to human language processing. In Proc. HLT, pages 262--267, San Francisco, 1994. Morgan Kaufmann.

....techniques and generative process techniques should be counterbalanced by the similarities that are brought to the fore when one re expresses rule based taggers as finite state transducers. Namely, HMM s can also be viewed as stochastic finite state transducers, as discussed by Pereira et al. [65]. This line of inquiry promises to give us a model of tagging (and partial parsing, as we shall see) of great generality, and is an area that will likely receive increasing attention. 2 Partial Parsing Let us turn now to parsing. Traditional parsers including standard stochastic parsers aim ....

....state sequence the automaton passes through. In this way it is possible to do stratal parsing with standard HMM training and recognition techniques. In more formal terms, we have turned each stratum automaton into a finitestate transducer, composed the transducers, and eliminated ffl transitions [48, 71, 65]. The only difference from standard transducer composition is that outputs at intermediate levels matter. The standard algorithms assume that states may be merged if doing so does not affect the relationship between the input and the final output. But in stratal parsing, we wish to keep states ....

Fernando C.N. Pereira, Michael Riley, and Richard W. Sproat. Weighted rational transductions and their application to human language processing. In Human Language Technology Workshop, pages 262--267, 1994.

....for reducing the sizes of language models while preserving overall recognition performance (accuracy and speed) Lacouture and De Mori [5] and Kenny et al. 4] discuss the size reduction of lexical trees in order to speed ASR systems, but they deal only with unweighted trees. Pereira et al. [7,8] discuss the use of weighted finite state automata (WFAs) to model human language in ASR systems. An advantage to this approach is that classical techniques for automata determinization and minimization can be extended to produce WFAs that are smaller yet formally equivalent to their inputs; i.e. ....

F. Pereira, M. Riley, and R. Sproat. Weighted rational transductions and their application to human language processing. In Proc. ARPA HLT, pages 249--54, 1994.

.... processing [22, 20, 21, 28, 30] A subclass of nite state machines, the weighted nite state automata (WFAs) has recently assumed new importance, because WFAs provide a powerful method for representing and manipulating models of human language in automatic speech recognition (ASR) systems [23, 24]. This new research direction also raises a number of challenging algorithmic questions [5] A weighted nite state automaton (WFA) is a nondeterministic nite automaton (NFA) A, that has both an alphabet symbol and a weight, from some set K, on each transition. Let R = K; 0; 1) be a ....

....Not all rational power series can be generated by deterministic WFAs. A determinization algorithm takes as input a WFA and produces a deterministic WFA that generates the same rational power series, if such a deterministic WFA exists. The importance of determinization to ASR is well established [20, 23, 24]. To the best of our knowledge, Mohri [20] presented the rst determinization procedure for WFAs, extending the seminal ideas of Cho rut [8, 9] and Weber and Klemm [31] regarding string to string transducers. Mohri gives a determinization procedure with three phases. First, A is converted to an ....

[Article contains additional citation context not shown here]

F. Pereira, M. Riley, and R. Sproat, Weighted rational transductions and their application to human language processing, in Proc. ARPA Human Language Technology Conf., 1994, pp. 249-54.

....cache. Both implementations were compiled with the Mongoose C compiler version 7.1. Table 1 lists our test graphs, which come from a variety of sources, along with their sizes. The ATIS, NAB, and PW graphs are derived from weighted finite state automata used in automatic speech recognition [17,18] by removing weights, labels, and multiple arcs. The different classes of graphs come from different speech tasks. The phone graphs represent assorted telephone calling patterns. The augmented binary graphs (AB1 21 T k M 1 M 2 M 2 k (a) 0 1 2 (b) 0 1 2 3 (c) Figure 15: a) T k is a ....

F. Pereira, M. Riley, and R. Sproat. Weighted rational transductions and their application to human language processing. In Proc. ARPA Human Language Technology Conf., pages 249--54, 1994.

....speech recognition, and modeling of biological sequences. The focus of this paper is on learning algorithms which have been developed for HMMs and many related models, such as hybrids of HMMs with artificial neural networks [1, 2, 3] Input Output HMMs [4, 5, 6, 7] weighted transducers [8, 9, 10, 11], variable length Markov models [12, 13] Markov switching models [14] and switching state space models [15, 16] Of course, there is a lot more litterature on HMMs and their applications than can be covered here, but this survey wants to be representative of the issues addressed here, mainly ....

.... in speech recognition HMMs, different sequences of speech units (corresponding to a subset of the possible state sequences) are associated with different weights (in fact the joint probability of these state sequences and the acoustic sequence) More generally, weighted acceptors and transducers [8, 9, 10, 11] can be used to assign a weight to a sequence (or a pair of input output sequences) Weighted acceptors and transducers are attractive in applications such as speech recognition and language processing because they can conveniently and uniformly represent and integrate different types of knowledge ....

[Article contains additional citation context not shown here]

F. Pereira, M. Riley, and R. Sproat, "Weighted rational transductions and their application to human language processing," in ARPA Natural Language Processing Workshop, 1994.

....contains the label of one of the possible categories, together with the penalty produced by the SDNN for that class label at that location. This graph is called the SDNN Output Graph. The second input graph to the transformer is a grammar transducer, more specifically a finite state transducer [Pereira et al. 1994], that encodes the relationship between input strings of class labels and corresponding output strings of recognized characters.The transducer is a weighted finite state machine (a graph) where each arc contains a pair of labels and possibly a penalty. Like a finite state machine, a transducer is ....

....al. 1995] for a review) However, there has been no proposal for a systematic approach to multi layer graph based trainable systems. The idea of transforming graphs into other graphs has received considerable interest in computer science, through the concept of weighted finite state transducers [Pereira et al. 1994]. Transducers have been applied to speech recognition [Pereira and Riley, 1997] and language translation [Mohri, 1997] and proposals have been made for handwriting recognition [Guyon et al. 1996] This line of work has been mainly focused on efficient search algorithms [Mohri and Riley, 1997] ....

[Article contains additional citation context not shown here]

Pereira, F., Riley, M., and Sproat, R. (1994). Weighted rational transductions and their application to human language processing. In ARPA Natural Language Processing workshop.

....contains the label of one of the possible categories, together with the penalty produced by the SDNN for that class label at that location. This graph is called the SDNN Output Graph. The second input graph to the transformer is a grammar transducer, more specifically a finite state transducer [86], that encodes the relationship between input strings of class labels and corresponding output strings of recognized characters. The transducer is a weighted finite state machine (a graph) where each arc contains a pair of labels and possibly a penalty. Like a finite state machine, a transducer is ....

....(see [38] for a review) However, there has been no proposal for a systematic approach to multi layer graph based trainable systems. The idea of transforming graphs into other graphs has received considerable interest in computer science, through the concept of weighted finite state transducers [86]. Transducers have been applied to speech recognition [100] and language translation [101] and proposals have been made for handwriting recognition [102] This line of work has been mainly focused on efficient search algorithms [103] and on the algebraic aspects of combining transducers and ....

[Article contains additional citation context not shown here]

F. Pereira, M. Riley, and R. Sproat, "Weighted rational transductions and their application to human language processing," in ARPA Natural Language Processing workshop, 1994.

....standard English corpora to train the main module while keeping the flexibility to modify the models for capitalization, numbers, symbols and proper nouns models, according to the needs of particular applications. The modules defined and composed use a special case of the approach of Peteira ct a [4]. This allows us to use graph search techniques, including beam search, without having to build explicitly the whole composed automaton. The classical measure of performance of language models is the cross entropy on a test set, which estimates the per character average length of the shortest ....

F. Pereira, M. Riley, and R. Sproat. Weighted rational transductions and their application to human language processing. In ARPA Natural Language Processing workshop, 1994.

....[11] and for performing other text analysis operations such as numeral expansion [9] this suggests a model of text analysis that is entirely based on regular relations. We present such a model below. More specifically we present a model of text analysis for TTS based on weighted FSTs (WFSTs) [7], which serves as the text analysis module of the multilingual Bell Labs TTS system. To date, the model has been applied to eight languages: Spanish, Italian, Romanian, French, German, Russian, Mandarin and Japanese. One property of this model that distinguishes it from most TTS text analyzers is ....

Fernando Pereira, Michael Riley, and Richard Sproat. Weighted rational transductions and their application to human language processing. In ARPA Workshop on Human Language Technology, pages 249--254. Advanced Research Projects Agency, March 8--11 1994.

....from the use of the VNSAs is threefold. First, the incorporation into a one pass Viterbi speech decoder is straightforward and efficient. Second, VNSAs can be exploited in a cascade of transducer compositions for speech processing (e.g. to include intra and interword phonotactic constraints) [8]. Thirdly, the VNSA is an effective method for training and implementing stochastic class based language models that outperform the standard n gram models in terms of perplexity and word accuracy [4] 5] 4. LANGUAGE MODELING WITH AUTOMATICALLY ACQUIRED PHRASES Traditionally, standard n gram ....

F. Pereira, M. Riley and R. Sproat, "Weighted rational transduction and their application to human language processing, " Proc. Workshop on Human Language Technology, pp. 249-254,Austin, 1994.

....and were chosen for their diversity: Alice in Wonderland, electronic mail, Unix man pages, and various Tex files. 4 Graph composition The overall system consists of several finite state automata, including the main VLMM and a proper noun VLMM. We use a special case of PFSA composition [3] to reconfigure flexibly the language model according to the needs of different applications, without having to retrain the core of the language model (Figure 4) For instance, the proper noun model can be replaced by a lexicon. Automata composition also allows us to use graph search techniques, ....

F. Pereira, M. Riley, and R. Sproat. Weighted rational transductions and their application to human language processing. In ARPA Natural Language Processing workshop, 1994.

No context found.

Fernando C. N. Pereira, Michael Riley, and Richard Sproat. 1994. Weighted rational transductions and their application to human language processing. In ARPA Workshop on Human Language Technology. Advanced Research Projects Agency.

....representation for all the inputs, outputs and domain information in speech recognition above the signal processing level. In particular, we use transducer composition to represent the combination of the various levels of acoustic, phonetic and linguistic information used in a recognizer [11]. For example, we may decompose a recognition task into a weighted acceptor O describing the acoustic observation sequence for the utterance to be recognized, a transducer A mapping acoustic observation sequences to context dependent phone sequences, a transducer C that converts between ....

Fernando C. N. Pereira, Michael Riley, and Richard W. Sproat, `Weighted rational transductions and their application to human language processing', in ARPA Workshop on Human Language Technology, (1994).

No context found.

Pereira, Fernando C. N., Michael Riley, and Richard W. Sproat. 1994. Weighted rational transductions and their application to human language processing. In ARPA Workshop on Human Language Technology. Morgan Kaufmann.

No context found.

Pereira, Fernando C. N., Michael Riley, and Richard W. Sproat. 1994. Weighted rational transductions and their application to human language processing. In ARPA Workshop on Human Language Technology. Morgan Kaufmann.

No context found.

F. Pereira, M. Riley, R. Sproat. Weighted rational transductions and their applications to human language processing. In Workshop on Human Language Technology Proc., pages 249--254, 1994.

No context found.

F. Pereira, M. Riley, and R. Sproat. Weighted rational transductions and their application to human language processing. In Proc. ARPA Human Language Technology Workshop '94, pages 249-254, Princeton, NJ, March 1994.

No context found.

Pereira, F., Riley, M. et Sproat, R. (1994). Weighted Rational Transductions and their Application to Human Language Processing. In Actes ARPA Workshop on Human Language Technology, pages 249--254.

No context found.

Fernando Pereira, Michael Riley, and Richard Sproat. 1994. Weighted rational transductions and their application to human language processing. In ARPA Workshop on Human Language Technology, pages 249--254. Advanced Research Projects Agency, March 8--11.

First 50 documents

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