Do we need bigram alignment models? On the effect of alignment quality on transduction accuracy in G2P

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1175–1185, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Do we need bigram alignment models? On the effect of alignment quality on transduction accuracy in G2P Steffen Eger Text Technology Lab Goethe University Frankfurt am Main steeger@em.uni-frankfurt.de Abstract We investigate the need for bigram align- ment models and the benefit of super- vised alignment techniques in grapheme- to-phoneme (G2P) conversion. Moreover, we quantitatively estimate the relation- ship between alignment quality and over- all G2P system performance. We find that, in English, bigram alignment models do perform better than unigram alignment models on the G2P task. Moreover, we find that supervised alignment techniques may perform considerably better than their unsupervised brethren and that few manu- ally aligned training pairs suffice for them to do so. Finally, we estimate a highly significant impact of alignment quality on overall G2P transcription performance and that this relationship is linear in nature. 1 Introduction Grapheme-to-phoneme (G2P) conversion is the problem of converting a string of letters into a string of phonetic symbols. Closely related to G2P are other string transduction problems in nat- ural language processing (NLP) such as transliter- ation (Sherif and Kondrak, 2007), lemmatization (Dreyer et al., 2008), and spelling error correc- tion (Brill and Moore, 2000). The classical learn- ing paradigm in each of these settings is to train a model on pairs of strings {(x, y)} and then to evaluate model performance on test data. While there are exceptions (e.g., (Rao et al., 2015)), most state-of-the-art modelings (e.g., (Jiampojamarn et al., 2007; Bisani and Ney, 2008; Jiampojamarn et al., 2008; Jiampojamarn et al., 2010; Novak et al., 2012)) view string transduction as a two-stage process in which string pairs (x, y) in the train- ing data are first aligned, and then a subsequent (e.g., sequence labeling) module is learned on the aligned data. ph oe n i x f i n I ks Table 1: Sample monotone many-to-many align- ment between x = phoenix and y = finIks. State-of-the-art alignments in G2P are charac- terized by the following properties: (i) Alignments are monotone in that the ordering of characters in input and output sequences is preserved by the alignments. Furthermore, they are many-to-many in the sense that sev- eral x sequence characters may be matched up with several y sequence characters as il- lustrated in Table 1. (ii) The alignment is a latent variable and learnt in an unsupervised manner from pairs of strings in the training data. (iii) The unsupervised alignment models are un- igram alignment models insofar as the over- all score that the alignment model assigns an alignment is the same for all orderings of the matched-up subsequences (context indepen- dence). To illustrate point (iii), consider, in the field of lemmatization, the case of aligning an inflected word form with the extended infinitive in German, such as absagt (‘rejects’) with abzusagen (‘to re- ject’). Critically, the insertion -zu- appears in in- fixal position …

Content truncated to 500 words