FonBund: A Library for Combining Cross-lingual Phonological Segment Data

FonBund: A Library for Combining Cross-lingual Phonological Segment Data Alexander Gutkin, Martin Jansche, Tatiana Merkulova Google London, United Kingdom {agutkin, mjansche, merkulova}@google.com Abstract We present an open-source library (FonBund) that provides a way of mapping sequences of arbitrary phonetic segments in International Phonetic Alphabet (IPA) into multiple articulatory feature representations. The library interfaces with several existing linguistic typology resources providing phonological segment inventories and their corresponding articulatory feature systems. Our first goal was to facili- tate the derivation of articulatory features without giving a special preference to any particular phonological segment inventory provided by freely available linguistic typology resources. The second goal was to build a very light-weight library that can be easily modified to support new phonological segment inventories. In order to support IPA segments that do not occur in the freely available resources, the library provides a simple configuration language for performing segment rewrites and adding custom segments with the corresponding feature structures. In addition to introducing the library and the corresponding linguistic resources, we also describe some of the practical uses of this library (multilingual speech synthesis) in the hope that this software will help facilitate multilingual speech research. Keywords: phonology, phonetic segments, software 1. Introduction Speech and language technology is currently only avail- able for a tiny fraction of the world’s languages. There has been a growing awareness of the importance of ad- dressing this disparity, especially in recent years. One of the outcomes of this realization is the appearance of sev- eral linguistic typology resources (Forkel, 2014) that aim to organize the world’s languages according to their struc- tural and functional features (O’Horan et al., 2016). One typical example is URIEL (Littell et al., 2017), a resource (and the corresponding software) that collates various fea- tures from various existing databases (Hammarstr¨om et al., 2015; Dryer and Haspelmath, 2013; Moran et al., 2014) that describe languages in terms of their phonological, lex- ical, morphosyntactic, phylogenetic and geographic distri- bution properties. Linguistic typology resources have been used in several ways to address the data scarcity problem for under- resourced languages. The first approach is multilingual joint learning where one hopes that training multiple lan- guages jointly will help with pooling the resources across languages and boost the performance on under-resourced language. Another popular approach is language trans- fer, where a resource-rich language is used to improve the performance of a resource-scarce language via model and data transfer (O’Horan et al., 2016). These ap- proaches were successfully used in several recent speech and language tasks such as grapheme-to-phoneme conver- sion (Deri and Knight, 2016; Peters et al., 2017), multi- lingual language modeling (Tsvetkov et al., 2016), text-to- speech (Tsvetkov, 2016), predicting missing language rep- resentation features (Malaviya et al., 2017) and name tag- ging (Zhang et al., 2017), among others (O’Horan et al., 2016). The focus of our work is on the linguistic resources that offer typology features describing the phonological struc- ture of the world’s languages. Such resources are extremely useful in multilingual speech research. Consider a …

Content truncated to 500 words