Morph-based speech recognition and modeling of out-of-vocabulary words across languages

Abstract

We explore the use of morph-based language models in large-vocabulary continuous-speech recognition systems across four so-called morphologically rich languages: Finnish, Estonian, Turkish, and Egyptian Colloquial Arabic. The morphs are subword units discovered in an unsupervised, data-driven way using the Morfessor algorithm. By estimating n-gram language models over sequences of morphs instead of words, the quality of the language model is improved through better vocabulary coverage and reduced data sparsity. Standard word models suffer from high out-ofvocabulary (OOV) rates, whereas the morph models can recognize previously unseen word forms by concatenating morphs. It is shown that the morph models do perform fairly well on OOVs without compromising the recognition accuracy on in-vocabulary words. The Arabic experiment constitutes the only exception since here the standard word model outperforms the morph model. Differences in the datasets and the amount of data are discussed as a plausible explanation.

THE RED SEQUENCE AT BIRTH IN THE GALAXY CLUSTER Cl J1449+0856 AT z = 2

“…To the best of our knowledge, this has not been tried before in the domain of text-entry. In the related domain of speech recognition, similar approaches have yielded good results [2] for agglutinative languages.…”

“…The task of the morph model is to assign a probability for each sequence of morphs in M(K) for the key sequence K. The probability of a sequence of s morphs (L 1 , ..., L s ) ∈ M(K) is given by the chain rule of probabilities in equation (2).…”

“…the vocabulary growth rate does not converge [2]. This means that OOV words present a serious problem for dictionary based approaches to predictive text entry of languages like Finnish and Turkish.…”

Unsupervised learning of agglutinated morphology using nested Pitman-Yor process based morpheme induction algorithm

“…Hammarström and Borin (2011) provide a detailed survey of the topic. Morfessor (Creutz and Lagus, 2002;Creutz and others, 2006;Creutz et al, 2007) based on Minimum Description Length principle is the reference model for highly inflecting languages, such as Finnish. Morfessor defines a model of lexicon and tries to find an optimum lexicon model using heuristic search procedure to achieve morphological segmentation.…”

Open-Lexicon Language Modeling Combining Word and Character Levels

“…Creutz [2] decomposes words into sub-word units called morphemes. The experiments were run for highly-inflecting languages, i.e.…”