Applications of Lexicographic Semirings to Problems in Speech and Language Processing

Abstract

This article 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).

Predicting Pronunciations with Syllabification and Stress with Recurrent Neural Networks

“…Russian, Spanish, Italian and Portuguese saw error rate reductions in the 50-60% range. While accuracy in Russian is slightly lower than Sproat et al [16] showed, our approach does not require any language-specific feature engineering.…”

“…Sproat et al [16] show that models can be trained to predict stress placement given spelling, following Dou et al [17] who report numbers on both stand-alone stress prediction as well as joint phoneme and stress prediction accuracy. Dou et al [17] report numbers over four European languages and use the CELEX2 [18] lexicon.…”

Neural Models of Text Normalization for Speech Applications

“…As a consequence, much of the subsequent work on applying machine learning to text normalization for speech applications focuses on specific semiotic classes, like letter sequences(Sproat and Hall 2014), abbreviations(Roark and Sproat 2014), or cardinal numbers(Gorman and Sproat 2016). 4 In fact, Kestrel(Ebden and Sproat 2014) uses a machine-learned morphosyntactic tagger for Russian.…”

Adaptation of Morph-Based Speech Recognition for Foreign Names and Acronyms

“…Pruning could still be done for acronym pronunciation variants. A classification algorithm based on Maximum Entropy-based rankers has previously been implemented to determine whether acronyms are pronounced as a word, letter sequence, or a mix of both [23].…”