A phoneme clustering algorithm based on the obligatory contour principle

Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017), pages 290–300, Vancouver, Canada, August 3 - August 4, 2017. c⃝2017 Association for Computational Linguistics A phoneme clustering algorithm based on the obligatory contour principle Mans Hulden Department of Linguistics University of Colorado mans.hulden@colorado.edu Abstract This paper explores a divisive hierarchi- cal clustering algorithm based on the well- known Obligatory Contour Principle in phonology. The purpose is twofold: to see if such an algorithm could be used for un- supervised classification of phonemes or graphemes in corpora, and to investigate whether this purported universal constraint really holds for several classes of phono- logical distinctive features. The algorithm achieves very high accuracies in an unsu- pervised setting of inferring a consonant- vowel distinction, and also has a strong tendency to detect coronal phonemes in an unsupervised fashion. Remaining classes, however, do not correspond as neatly to phonological distinctive feature splits. While the results offer only mixed support for a universal Obligatory Contour Princi- ple, the algorithm can be very useful for many NLP tasks due to the high accuracy in revealing consonant/vowel/coronal dis- tinctions. 1 Introduction1 It has long been noted in phonology that there seems to be a universal cross-linguistic tendency to avoid redundancy or repetition of similar speech features within a word or morpheme, especially if the phonemes are adjacent to one another. Many different names are given to variants of this gen- eral phenomenon in the linguistic literature: “iden- tity avoidance” (Yip, 1998), “similar place avoid- ance” (Pozdniakov and Segerer, 2007), “oblig- atory contour principle” (OCP) (Leben, 1973), and “dissimilation” (Hempl, 1893). Some special 1All code data sets used are available at https:// github.com/cvocp/cvocp cases such as haplology (avoidance of adjacent identical syllables) also fall in this general cate- gory of avoiding repetition along some dimension. The general phenomenon itself is supported by robust, although inconsistent, evidence across a number of languages. An early example is the observation of Spitta-Bey (1880),2 that the Ara- bic language tends to favor combination of con- sonant segments (phonemes) in morphemes that have different places of articulation; this was also later pointed out by Greenberg (1950) and those Semitic root outliers that deviate from this pat- tern were analyzed in depth in Frajzyngier (1979). In Proto-Indo-European (PIE) roots, which are mostly structured CVC, stop-V-stop combinations have been found to be statistically underrepre- sented (Iverson and Salmons, 1992). That is, PIE seems to obey a cross-linguistic constraint that dis- favors two similar consonants in a root. Another specific example comes from Japanese, where the phenomenon called Lyman’s law—which effec- tively says that a morpheme may consist of max- imally one voiced obstruent—can also be inter- preted as avoidance (Itˆo and Mester, 1986). In light of such evidence, proposals have been put forth to define the concept of phoneme by distributional properties alone as opposed to the prevalent distinctive feature systems which are largely based on articulatory features (Fischer- Jørgensen, 1952). Elsewhere, after finding a sta- tistical tendency to avoid similar place …

Content truncated to 500 words