Abhirut Gupta, Ananya B. Sai, Richard Sproat, Yuri Vasilevski, James Ren, Ambarish Jash, Sukhdeep Sodhi, Aravindan Raghuveer. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Abstract If one sees the place name Houston Mercer Dog Run in New York, how does one know how to pronounce it? Assuming one knows that Houston in New York is pronounced /ˈhaʊstən/ and not like the Texas city (/ˈhjuːstən/), then one can probably guess that /ˈhaʊstən/ is also used in the name of the dog park. We present a novel architecture that learns to use the pronunciations of neighboring names in order to guess the pronunciation of a given target feature. Applied to Japanese place names, we demonstrate the utility of the model to finding and proposing corrections for errors in Google Maps. To demonstrate the utility of this approach to structurally similar problems, we also report on an application to a totally different task: Cognate reflex prediction in comparative historical linguistics. A version of the code has been open-sourced.1

Abstract In a recent position paper, Turing Award Winners Yoshua Bengio, Geoffrey Hinton, and Yann LeCun make the case that symbolic methods are not needed in AI and that, while there are still many issues to be resolved, AI will be solved using purely neural methods. In this piece I issue a challenge: Demonstrate that a purely neural approach to the problem of text normalization is possible. Various groups have tried, but so far nobody has eliminated the problem of unrecoverable errors, errors where, due to insufficient training data or faulty generalization, the system substitutes some other reading for the correct one. Solutions have been proposed that involve a marriage of traditional finite-state methods with neural models, but thus far nobody has shown that the problem can be solved using neural methods alone. Though text normalization is hardly an “exciting” problem, I argue that until one can solve “boring” problems like that using purely AI methods, one cannot claim that AI is a success.

Taxonomies of writing systems since Gelb (1952) have classified systems based on what the written symbols represent: if they represent words or morphemes, they are logographic; if syllables, syllabic; if segments, alphabetic; and so forth. Sproat (2000) and Rogers (2005) broke with tradition by splitting the logographic and phonographic aspects into two dimensions, with logography being graded rather than a categorical distinction. A system could be syllabic, and highly logographic; or alphabetic, and mostly non-logographic. This accords better with how writing systems actually work, but neither author proposed a method for measuring logography. In this article we propose a novel measure of the degree of logography that uses an attention-based sequence-to-sequence model trained to predict the spelling of a token from its pronunciation in context. In an ideal phonographic system, the model should need to attend to only the current token in order to compute how to spell it, and this would show in the attention matrix activations. In contrast, with a logographic system, where a given pronunciation might correspond to several different spellings, the model would need to attend to a broader context. The ratio of the activation outside the token and the total activation forms the basis of our measure. We compare this with a simple lexical measure, and an entropic measure, as well as several other neural models, and argue that on balance our attention-based measure accords best with intuition about how logographic various systems are. Our work provides the first quantifiable measure of the notion of logography that accords with linguistic intuition and, we argue, provides better insight into what this notion means.

Breaking domain names such as openresearch into component words open and research is important for applications like Text-to-Speech synthesis and web search.We link this problem to the classic problem of Chinese word segmentation and show the effectiveness of a tagging model based on Recurrent Neural Networks (RNNs) using characters as input.To compensate for the lack of training data, we propose a pre-training method on concatenated entity names in a large knowledge database.Pre-training improves the model by 33% and brings the sequence accuracy to 85%.

Machine learning, including neural network techniques, have been applied to virtually every domain in natural language processing. One problem that has been somewhat resistant to effective machine learning solutions is text normalization for speech applications such as text-to-speech synthesis (TTS). In this application, one must decide, for example, that 123 is verbalized as one hundred twenty three in 123 pages but as one twenty three in 123 King Ave. For this task, state-of-the-art industrial systems depend heavily on hand-written language-specific grammars. We propose neural network models that treat text normalization for TTS as a sequence-to-sequence problem, in which the input is a text token in context, and the output is the verbalization of that token. We find that the most effective model, in accuracy and efficiency, is one where the sentential context is computed once and the results of that computation are combined with the computation of each token in sequence to compute the verbalization. This model allows for a great deal of flexibility in terms of representing the context, and also allows us to integrate tagging and segmentation into the process. These models perform very well overall, but occasionally they will predict wildly inappropriate verbalizations, such as reading 3 cm as three kilometers. Although rare, such verbalizations are a major issue for TTS applications. We thus use finite-state covering grammars to guide the neural models, either during training and decoding, or just during decoding, away from such “unrecoverable” errors. Such grammars can largely be learned from data.

Attention-based sequence-to-sequence neural network models learn to jointly align and translate. The quadratic-time attention mechanism is powerful as it is capable of handling arbitrary long-distance reordering, but computationally expensive. In this paper, towards making neural translation both accurate and efficient, we follow the traditional pre-reordering approach to decouple reordering from translation. We add a reordering RNN that shares the input encoder with the decoder. The RNNs are trained jointly with a multi-task loss function and applied sequentially at inference time. The task of the reordering model is to predict the permutation of the input words following the target language word order. After reordering, the attention in the decoder becomes more peaked and monotonic. For reordering, we adopt the Inversion Transduction Grammars (ITG) and propose a transition system to parse input to trees for reordering. We harness the ITG transition system with RNN. With the modeling power of RNN, we achieve superior reordering accuracy without any feature engineering. In experiments, we apply the model to the task of text normalization. Compared to a strong baseline of attention-based RNN, our ITG RNN re-ordering model can reach the same reordering accuracy with only 1/10 of the training data and is 2.5x faster in decoding.

Text normalization is the task of converting from a written representation into a representation of how the text is to be spoken. For most real-world speech applications, the text normalization engine is developed mostly by hand. For example, a hand-built grammar may be used to enumerate possible ways to say a given token in a given language, and a statistical model used to select the most appropriate verbalizations in context. We examine the tradeoffs associated with using more or less language-specific knowledge for text normalization. In the most data-rich scenario, we have access to a carefully constructed hand-built normalization grammar that for any given token will produce a lattice of all possible verbalizations for that token. We assume a parallel corpus of aligned written-spoken utterances. As a substitute for the hand-built grammar, we consider a language-universal normalization covering grammar, where the developer merely needs to provide a set of lexical items particular to the language. As a substitute for the aligned corpus, we consider a scenario where one only has the spoken side, and the corresponding written side is “hallucinated” by composing the spoken side with the inverted normalization grammar. We report performance of the above scenarios on experiments with English and Russian.

We propose two models for verbalizing numbers, a key component in speech recognition and synthesis systems. The first model uses an end-to-end recurrent neural network. The second model, drawing inspiration from the linguistics literature, uses finite-state transducers constructed with a minimal amount of training data. While both models achieve near-perfect performance, the latter model can be trained using several orders of magnitude less data than the former, making it particularly useful for low-resource languages.

Abstract This paper describes the Kestrel text normalization system, a component of the Google text-to-speech synthesis (TTS) system. At the core of Kestrel are text-normalization grammars that are compiled into libraries of weighted finite-state transducers (WFSTs). While the use of WFSTs for text normalization is itself not new, Kestrel differs from previous systems in its separation of the initialtokenization and classificationphase of analysis fromverbalization. Input text is first tokenized and different tokens classified using WFSTs. As part of the classification, detectedsemiotic classes– expressions such as currency amounts, dates, times, measure phases, are parsed into protocol buffers ( https://code.google.com/p/protobuf/ ). The protocol buffers are then verbalized, with possible reordering of the elements, again using WFSTs. This paper describes the architecture of Kestrel, the protocol buffer representations of semiotic classes, and presents some examples of grammars for various languages. We also discuss applications and deployments of Kestrel as part of the Google TTS system, which runs on both server and client side on multiple devices, and is used daily by millions of people in nineteen languages and counting.

No abstract available.

Deficits in semantic and pragmatic expression are among the hallmark linguistic features of autism. Recent work in deriving computational correlates of clinical spoken language measures has demonstrated the utility of automated linguistic analysis for characterizing the language of children with autism. Most of this research, however, has focused either on young children still acquiring language or on small populations covering a wide age range. In this paper, we extract numerous linguistic features from narratives produced by two groups of children with and without autism from two narrow age ranges. We find that although many differences between diagnostic groups remain constant with age, certain pragmatic measures, particularly the ability to remain on topic and avoid digressions, seem to improve. These results confirm findings reported in the psychology literature while underscoring the need for careful consideration of the age range of the population under investigation when performing clinically oriented computational analysis of spoken language.

This paper explores lexicographic semirings and their application to problems in speech and language processing. Specifically, we present two instantiations of binary lexicographic semirings, one involving a pair of tropical weights, and the other a tropical weight paired with a novel string semiring we term the categorial semiring. The first of these is used to yield an exact encoding of backoff models with epsilon transitions. This lexicographic language model semiring allows for off-line optimization of exact models represented as large weighted finite-state transducers in contrast to implicit (on-line) failure transition representations. We present empirical results demonstrating that, even in simple intersection scenarios amenable to the use of failure transitions, the use of the more powerful lexicographic semiring is competitive in terms of time of intersection. The second of these lexicographic semirings is applied to the problem of extracting, from a lattice of word sequences tagged for part of speech, only the single best-scoring part of speech tagging for each word sequence. We do this by incorporating the tags as a categorial weight in the second component of a 〈Tropical, Categorial〉 lexicographic semiring, determinizing the resulting word lattice acceptor in that semiring, and then mapping the tags back as output labels of the word lattice transducer. We compare our approach to a competing method due to Povey et al. (2012).

No abstract available.

We explore a model of stress prediction in Russian using a combination of local contextual features and linguisticallymotivated features associated with the word's stem and suffix.We frame this as a ranking problem, where the objective is to rank the pronunciation with the correct stress above those with incorrect stress.We train our models using a simple Maximum Entropy ranking framework allowing for efficient prediction.An empirical evaluation shows that a model combining the local contextual features and the linguistically-motivated non-local features performs best in identifying both primary and secondary stress.

No abstract available.

No abstract available.

Speech and language processing systems routinely face the need to apply finite state operations (e.g., POS tagging) on results from intermediate stages (e.g., ASR output) that are naturally represented in a compact lattice form. Currently, such needs are met by converting the lattices into linear sequences (n-best scoring sequences) before and after applying the finite state operations. In this paper, we eliminate the need for this unnecessary conversion by addressing the problem of picking only the single-best scoring output labels for every input sequence. For this purpose, we define a categorial semiring that allows determinzation over strings and incorporate it into a 〈Tropical, Categorial〉 lexicographic semiring. Through examples and empirical evaluations we show how determinization in this lexicographic semiring produces the desired output. The proposed solution is general in nature and can be applied to multi-tape weighted transducers that arise in many applications.

In this work, we study a new text mining problem of discovering named entities with temporally correlated bursts of mention counts in multiple multilingual Web news streams. Mining named entities with temporally correlated bursts of mention counts in multilingual text streams has many interesting and important applications, such as identification of the latent events, attracting the attention of on-line media in different countries, and valuable linguistic knowledge in the form of transliterations. While mining "bursty" terms in a single text stream has been studied before, the problem of detecting terms with temporally correlated bursts in multilingual Web streams raises two new challenges: (i) correlated terms in multiple streams may have bursts that are of different orders of magnitude in their intensity and (ii) bursts of correlated terms may be separated by time gaps. We propose a two-stage method for mining items with temporally correlated bursts from multiple data streams, which addresses both challenges. In the first stage of the method, the temporal behavior of different entities is normalized by modeling them with the Markov-Modulated Poisson Process. In the second stage, a dynamic programming algorithm is used to discover correlated bursts of different items, that can be potentially separated by time gaps. We evaluated our method with the task of discovering transliterations of named entities from multilingual Web news streams. Experimental results indicate that our method can not only effectively discover named entities with correlated bursts in multilingual Web news streams, but also outperforms two state-of-the-art baseline methods for unsupervised discovery of transliterations in static text collections.

We present a pronunciation error detection method for second language learners of English (L2 learners). The method is a combination of confidence scoring and landmark-based Support Vector Machines (SVMs). Landmark-based SVMs were implemented to specialize the method for the specific phonemes with which L2 learners make frequent errors. The method was trained for the difficult phonemes for Korean learners and tested on intermediate Korean learners. In the data where distortion errors (non-phonemic errors) occupied high proportion, SVM method achieved significantly higher F-score (0.67) than confidence scoring (0.60). However, the combination of two methods without the appropriate training data did not lead to improvement. Even for intermediate learners, a high proportion of errors (40%) was related to these difficult phonemes. Therefore, the method specialized for these phonemes will be beneficial for both beginners and intermediate learners. Index Terms: automated pronunciation error detection, computer aided pronunciation training system, confidence score,

Most areas of natural language processing today make heavy use of automatic inference from large corpora. One exception is text-normalization for such applications as text-to-speech synthesis, where it is still the norm to build grammars by hand for such tasks as handling abbreviations or the expansion of digit sequences into number names. One reason for this, apart from the general lack of interest in text normalization, has been the lack of annotated data. For many languages, however, there is abundant unannotated data that can be brought to bear on these problems. This paper reports on the inference of number-name expansion in Russian, a particularly difficult language due to its complex inflectional system. A database of several million spelled-out number names was collected from the web and mapped to digit strings using an overgenerating number-name grammar. The same overgenerating number-name grammar can be used to produce candidate expansions into number names, which are then scored using a language model trained on the web data. Our results suggest that it is possible to infer expansion modules for very complex number name systems, from unannotated data, and using a minimum of hand-compiled seed data.

People have used the existence of quasi-Zipfian distributions in symbol systems to argue for their status as writing; such claims figure in the work of Rao and colleagues.But because it has been long known that Zipfian distributions hold of many things besides language, such arguments are easy to dismiss.

In the last issue of this journal, I presented a piece that called into question some of the techniques reported in two papers in high-profile journals that purported to provide statistical evidence for the linguistic status of some ancient symbol systems (Sproat 2010a). Not surprisingly, the authors of those two papers took issue with a number of my claims, and have requested the opportunity to respond. The two responses, taken together, are rather lengthy and as a result it is not possible, given limitations of space, for me to address each and every one of their criticisms. I will therefore focus on what I consider to be the most important objections.

In this study, we present a pronunciation scoring method for second language learners of English (hereafter, L2 learners). This study presents a method using both confidence scoring and classifiers. Classifiers have an advantage over confidence scoring for specialization in the specific phonemes where L2 learners make frequent errors. Classifiers (Landmark-based Support Vector Machines) were trained in order to distinguish L2 phonemes from their frequent substitution patterns. In this study, the method was evaluated on the specific English phonemes where L2 English learners make frequent errors. The results suggest that the automated pronunciation scoring method can be improved consistently by combining the two methods.

For readers of traditional textbooks such as that of Partee, ter Meulen, and Wall (1990), the term 'mathematical linguistics' denotes a rather narrowly circumscribed set of issues including automata theory, set theory, and lambda calculus, with maybe a little formal language theory thrown in.Kornai's contribution is refreshingly different in that he treats, in this relatively compact volume, practically all areas of linguistics, phonetics, and speech and language processing.Kornai's motivation for writing this book is to present "a single entry point to the central methods and concepts of linguistics that are made largely inaccessible to the mathematician, computer scientist, or engineer by the surprisingly adversarial style of argumentation . . .and the proliferation of unmotivated notation and formalism . . .all too often encountered in research papers and monographs in the humanities" (page viii).There is no question that much of what passes for rigor (mathematical and scientific) in linguistics is a joke and that there is clearly a need for any work that can place the field on a more solid footing.It also seems likely that Kornai is the only person who could have written this book.The book is divided into ten chapters, including a short introductory chapter, which lays the groundwork and identifies the potential audience, and a concluding chapter where Kornai reveals his own views on what is important in the field, which in the interests of balance he has largely suppressed throughout the book.Chapter 2 is also introductory in that it presents basic concepts of generation (via a ruleset), axioms, and string rewriting.The main chapters (3-9) deal with a variety of topic areas relating to language and speech, starting with phonology in Chapter 3.This chapter introduces the notion of phonemes, distinctive features, autosegmental phonology, and computation using finite automata.Kornai offers many details that are of course lacking in most linguistic treatments, such as a proof that the number of well-formed association lines between two tiers of length n is asymptotically (6 + 4 √ 2) n .Chapter 4 deals with morphology, which for Kornai includes not only word formation, but also prosody (including stress assignment and moraic structure), as well as Optimality Theory and Zipf's law.The fifth chapter treats syntax, including categorial grammar, phrase structure, dependency frameworks, valency, and weighted models of grammar, ending with a discussion of weighted finite automata and hidden Markov models.In the context of weighted models, Kornai implies that Chomsky's original notion of degree of grammaticality fits naturally as an instance of a weighted model with a particular semiring; of course, exactly what the ⊕ and ⊗ operators of that semiring map to remain to

Previous work on text mining has almost exclusively focused on a single stream. However, we often have available multiple text streams indexed by the same set of time points (called coordinated text streams), which offer new opportunities for text mining. For example, when a major event happens, all the news articles published by different agencies in different languages tend to cover the same event for a certain period, exhibiting a correlated bursty topic pattern in all the news article streams. In general, mining correlated bursty topic patterns from coordinated text streams can reveal interesting latent associations or events behind these streams. In this paper, we define and study this novel text mining problem. We propose a general probabilistic algorithm which can effectively discover correlated bursty patterns and their bursty periods across text streams even if the streams have completely different vocabularies (e.g., English vs Chinese). Evaluation of the proposed method on a news data set and a literature data set shows that it can effectively discover quite meaningful topic patterns from both data sets: the patterns discovered from the news data set accurately reveal the major common events covered in the two streams of news articles (in English and Chinese, respectively), while the patterns discovered from two database publication streams match well with the major research paradigm shifts in database research. Since the proposed method is general and does not require the streams to share vocabulary, it can be applied to any coordinated text streams to discover correlated topic patterns that burst in multiple streams in the same period.

In this paper we investigate Chinese-English name transliteration using comparable corpora, corpora where texts in the two languages deal in some of the same topics --- and therefore share references to named entities --- but are not translations of each other. We present two distinct methods for transliteration, one approach using phonetic transliteration, and the second using the temporal distribution of candidate pairs. Each of these approaches works quite well, but by combining the approaches one can achieve even better results. We then propose a novel score propagation method that utilizes the co-occurrence of transliteration pairs within document pairs. This propagation method achieves further improvement over the best results from the previous step.

Whereas experimental studies on emotional speech often control for neutral semantics, speech in naturalistic speech corpora is characterized by contextual cues and non-neutral semantic content. Moreover, the target emotion of an utterance is generally unknown and must be inferred by the listener. Within the context of having child-directed expressive text-to-speech synthesis as goal, we describe a perceptual study based on an expressive spoken corpus of children’s stories with unknown emotional targets, and report on interannotator agreement in a forced-choice discrimination task. Moreover, a threshold of high agreement was used to establish subsets of confident exemplar utterances for emotional classes, comprising 35% of the initial corpus. The exemplars were clustered based on the differences from the default mean neutral for 11 global acoustic features, yielding clusters cutting across emotion boundaries, some of which reflected arousal levels, with the neutral exemplars showing particularly complex distributions. Moreover, the mean features for four emotional exemplar categories were contrasted against the default, finding both expected and contradictory tendencies, compared to previous reports. The results indicate that semantic and prosodic cues collaborate to express and reinforce emotional contents, while emotional sequencing seems likely to be another factor which contributes to emotional perception in this domain.

As speech recognition systems are used in ever more applications, it is crucial for the systems to be able to deal with accented speakers. Various techniques, such as acoustic model adaptation and pronunciation adaptation, have been reported to improve the recognition of non-native or accented speech. In this paper, we propose a new approach that combines accent detection, accent discriminative acoustic features, acoustic adaptation and model selection for accented Chinese speech recognition. Experimental results show that this approach can improve the recognition of accented speech. 1.

A transcription system that requires accurate general name transcription is faced with the problem of covering the large number of names it may encounter, Without any prior knowledge, this requires a large increase in the size and complexity of the system due to the expansion of the lexicon. Furthermore, this increase will adversely affect the system performance due to the increased confusability. Here we propose a method that uses meta-data, available at runtime to ensure better name coverage without significantly increasing the system complexity. We tested this approach on a voicemail transcription task and assumed meta-data to be available in the form of a caller ID string (as it would show up on a caller ID enabled telephone) and the name of the mailbox owner. Networks representing possible spoken realization of those names are generated at runtime and included in the network of the decoder. The decoder network is built at training time using a class-dependent language model, with caller and mailbox name instances modeled as class tokens. The class tokens are replaced at test time with the name networks built from the meta-data. The proposed algorithm showed a reduction in the error rate of name tokens of 22.1%.

Choosing appropriate voices for synthesizing children’s stories requires text analysis techniques that can identify which portions of the text should be read by which speakers. Our work presents techniques to take raw text stories and automatically identify the quoted speech, identify the characters within the stories and assign characters to each quote. The resulting marked-up story may then be rendered with a standard speech synthesizer with appropriate voices for the characters. This paper presents each of the basic stages in identification, and the algorithms, both rule-driven and data-driven, used to achieve this. A variety of story texts are used to test our system. Results are presented with a discussion of the limitations and recommendations on how to improve speaker assignment in further texts.

There has been a great deal of work over the past decade on inferring semantic information from text corpora. This paper is another instance of this kind of work, but is also slightly different in that we are interested not in extracting semantic information per se, but rather real-world knowledge. In particular, given a description of a particular action --- e.g. John was eating breakfast --- we want to know where John is likely to be, what time of day it is, and so forth. Humans on hearing this sentence would form a mental image that makes a lot of inferences about the environment in which this action occurs: they would probably imagine someone in their kitchen in the morning, perhaps in their dining room, seated at a table, eating a meal.We propose a method that makes use of Dunning's likelihood ratios to extract from text corpora strong associations between particular actions and locations or times when those actions occur. We also present an evaluation of the method. The context of this work is a text-to-scene conversion system called WordsEye, where in order to depict an action such as John was eating breakfast, it is desirable to make reasonable inferences about where and when that action is taking place so that the resulting picture is a reasonable match to one's mental image of the action.

Natural language is an easy and effective medium for describing visual ideas and mental images. Thus, we foresee the emergence of language-based 3D scene generation systems to let ordinary users quickly create 3D scenes without having to learn special software, acquire artistic skills, or even touch a desktop window-oriented interface. WordsEye is such a system for automatically converting text into representative 3D scenes. WordsEye relies on a large database of 3D models and poses to depict entities and actions. Every 3D model can have associated shape displacements, spatial tags, and functional properties to be used in the depiction process. We describe the linguistic analysis and depiction techniques used by WordsEye along with some general strategies by which more abstract concepts are made depictable.

We propose a system for automatically determining boundaries between phonetic segments in a speechwave given a phonetic transcription: automatic segmentation. The system uses edge detectors that are applied to various speech representations; both are optimized for each diphone or diphone class. Output from these detectors, which contains spuriously detected edges, is then combined with alternative pronunciations generated via rules from the canonical pronunciation. The #nal output is generated with lowest-cost path algorithms applied to #nite state transducers. 1. INTRODUCTION Automatic segmentation is critical both for speech research and for speech technologies that rely on segmented speech corpora for training or construction purposes. For example, in text-to-speech synthesis #TTS# segmented corpora are used for the construction of intonation, duration, and synthesis components #4#. The standard approach to automated segmentation is to adapt an automatic speech recognition #ASR#...

The signature block is a common structured component found in email messages. Accurate identification and analysis of signature blocks is important in many multimedia messaging and information retrieval applications such as email text-to-speech rendering, automatic construction of personal address databases, and interactive message retrieval. It is also a very challenging task, because signature blocks often appear in complex two-dimensional layouts which are guided only by loose conventions. Traditional text analysis methods designed to deal with sequential text cannot handle two-dimensional structures, while the highly unconstrained nature of signature blocks makes the application of two-dimensional grammars very difficult. In this article, we describe an algorithm for signature block analysis which combines two-dimensional structural segmentation with one-dimensional grammatical constraints. The information obtained from both layout and linguistic analysis is integrated in the form of weighted finite-state transducers. The algorithm is currently implemented as a component in a preprocessing system for email text-to-speech rendering.

The signature block is a common structured component found in e-mail messages. Accurate identification and analysis of signature blocks are important in many multimedia messaging and information retrieval applications such as e-mail text-to-speech rendering. Traditional text analysis methods designed to deal with sequential text cannot handle 2D structures, while the highly unconstrained nature of signature blocks makes the application of 2D grammars very difficult. In this paper we describe an algorithm for signature block analysis which combines 2D structural segmentation with 1D grammatical constraints. The information obtained from both geometrical and linguistic analysis are integrated in a form of weighted finite state transducers, and the final solution is the optimal interpretation under both constraints.

Currently, speech synthesizers are controlled by a multitude of proprietary tag sets. These tag sets vary substantially across synthesizers and are an inhibitor to the adoption of speech synthesis technology by developers. SABLE is an XML/SGML-based markup scheme for text-to-speech synthesis, developed to address the need for a common TTS control paradigm. This paper presents an overview of the SABLE specification, and provides links to sites where further information on SABLE can be accessed.

Most work on sense disambiguation presumes that one knows beforehand -e.g. from a thesaurus -a set of polysemous terms.But published lists invariably give only partial coverage.For example, the English word tan has several obvious senses, but one may overlook the abbreviation for tangent.In this paper, we present an algorithm for identifying interesting polysemous terms and measuring their degree of polysemy, given an unlabeled corpus.The algorithm involves: (i) collecting all terms within a -term window of the target term; (ii) computing the inter-term distances of the contextual terms, and reducing the multi-dimensional distance space to two dimensions using standard methods; (iii) converting the two-dimensional representation into radial coordinates and using isotonic/antitonic regression to compute the degree to which the distribution deviates from a single-peak model.The amount of deviation is the proposed polysemy index.

E-mail reading is one of the most important commercial applications of text-to-speech synthesis (TTS). Yet e-mail is one of the most difficult types of text to deal with, since it is both highly structured -frequently containing elements such as tables, signatures, "artwork" and quotations from previous messages; and at the same time often lacks any reliable unambiguous indicators for such structure. This paper describes Emu, an e-mail mark-up and rendering program that preprocesses e-mail for TTS. We discuss algorithms for detecting regions of interest in the input text; for "normalizing" the input; and for actually rendering the input through the Bell Labs TTS system.

In this paper we present an overview of the German version of the Bell Labs text-to-speech system, a high-quality concatenative synthesis system with extensive text analysis capabilities. We discuss problems of text analysis, and our solutions to these problems, including: the integration of text normalization tasks into linguistic text analysis; the capability to morphologically analyze compounds and unseen words; name analysis and pronunciation. We briefly describe the prosodic components of the text-to-speech system and their underlying duration and intonation models. Finally, the phonetically motivated structure of the acoustic inventory is presented.

This paper describes the Bell Labs Russian text-to-speech system, a concatenative system with extensive text-analysis capabilities. The construction of Russian-specific modules will be discussed, including the text-analysis module, the acoustic inventory, the duration module, and the intonation module.

Context-dependent rewrite rules are used in many areas of natural language and speech processing. Work in computational phonology has demonstrated that, given certain conditions, such rewrite rules can be represented as finite-state transducers (FSTs). We describe a new algorithm for compiling rewrite rules into FSTs. We show the algorithm to be simpler and more efficient than existing algorithms. Further, many of our applications demand the ability to compile weighted rules into weighted FSTs, transducers generalized by providing transitions with weights. We have extended the algorithm to allow for this.

We report on a method for compiling decision trees into weighted finite-state transducers. The key assumptions are that the tree predictions specify how to rewrite symbols from an input string, and the decision at each tree node is stateable in terms of regular expressions on the input string. Each leaf node can then be treated as a separate rule where the left and right contexts are constructable from the decisions made traversing the tree from the root to the leaf. These rules are compiled into transducers using the weighted rewite-rule rule-compilation algorithm described in (Mohri and Sproat, 1996).

We present a model of text analysis for text-to-speech (TTS) synthesis based on (weighted) finite-state transducers, which serves as the text-analysis module of the multilingual Bell Labs TTS system. The transducers are constructed using a lexical toolkit that allows declarative descriptions of lexicons, morphological rules, numeral-expansion rules, and phonological rules, inter alia. To date, the model has been applied to eight languages: Spanish, Italian, Romanian, French, German, Russian, Mandarin and Japanese.

We present a model of text analysis for text-to-speech (TTS) synthesis based on (weighted) finite state transducers, which serves as the text analysis module of the multilingual Bell Labs TTS system. The transducers are constructed using a lexical toolkit that allows declarative descriptions of lexicons, morphological rules, numeral-expansion rules, and phonological rules, inter alia. To date, the model has been applied to eight languages: Spanish, Italian, Romanian, French, German, Russian, Mandarin and Japanese.

We present a stochastic finite-state model for segmenting Chinese text into dictionary entries and productively derived words, and providing pronunciations for these words; the method incorporates a class-based model in its treatment of personal names. We also evaluate the system's performance, taking into account the fact that people often do not agree on a single segmentation.

A problem with many speech understanding systems is that grammars that are more suitable for representing the relation between sentences and their meanings, such as context free grammars (CFGs) and augmented phrase structure grammars (APSGs), are computationally very demanding. On the other hand, finite state grammars are efficient, but cannot represent directly the sentence-meaning relation. The authors describe how speech recognition and language analysis can be tightly coupled by developing an APSG for the analysis component and deriving automatically from it a finite-state approximation that is used as the recognition language model. Using this technique, the authors have built an efficient translation system that is fast compared to others with comparably sized language models.>

We present a model of morphological processing which directly encodes prosodic constituency, a notion which is clearly crucial in many widespread morphological processes. The model has been implemented for the Australian language Warlpiri and has been successfully interfaced with a syntactic parser for that language (Brunson, 1986). We contrast our approach with approaches to morphological parsing in the KIMMO framework.

We describe a program for assigning correct stress contours to nominals in English.It makes use of idiosyncratic knowledge about the stress behavior of various nominal types and general knowledge about English stress rules.We have also investigated the related issue of parsing complex nominals in English.The importance of this work and related research to the problem of text-to-speech is 'discussed.