Contrasting Vertical and Horizontal Transmission of Typological Features

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 836–846, Osaka, Japan, December 11-17 2016. Contrasting Vertical and Horizontal Transmission of Typological Features Kenji Yamauchi Yugo Murawaki Graduate School of Informatics, Kyoto University Yoshida-honmachi, Sakyo-ku, Kyoto, 606-8501, Japan yamauchi@nlp.ist.i.kyoto-u.ac.jp murawaki@i.kyoto-u.ac.jp Abstract Linguistic typology provides features that have a potential of uncovering deep phylogenetic re- lations among the world’s languages. One of the key challenges in using typological features for phylogenetic inference is that horizontal (spatial) transmission obscures vertical (phylogenetic) signals. In this paper, we characterize typological features with respect to the relative strength of vertical and horizontal transmission. To do this, we first construct (1) a spatial neighbor graph of languages and (2) a phylogenetic neighbor graph by collapsing known language families. We then develop an autologistic model that predicts a feature’s distribution from these two graphs. In the experiments, we managed to separate vertically and/or horizontally stable features from unstable ones, and the results are largely consistent with previous findings. 1 Introduction Centuries of research in historical linguistics have identified groups of languages deriving from single common ancestors. Each of these groups, called a language family, is organized as a tree that reflects its evolutionary history. Such language families include Indo-European, Austronesian and Bantu languages. Despite the huge success, historical linguistics fails to link some languages with others, and hence they are called language isolates. For example, Basque, Burushaski and Japanese have no established relatives. Although there are several attempts to uncover deep phylogenetic relations, in which the results are often represented as macrofamilies, they remain controversial (Dolgopolsky, 1998; Greenberg, 2000). We argue that the limitation of the mainstream approach is that it relies on lexical traits. Language families are established by demonstrating that its member languages share cognates, or words that have a common etymological origin.1 Lexical data are also the main target of modern statistical methods that are typically used to date the common ancestor (Swadesh, 1971; Gray and Atkinson, 2003). Language isolates are called so exactly because they lack reliable cognates. For this reason, we follow a different line to research that makes use of typological data for phyloge- netic inference (Tsunoda et al., 1995; Dunn et al., 2005; Teh et al., 2008; Longobardi and Guardiano, 2009; Murawaki, 2015). Linguistic typology is a cross-linguistic study that classifies the world’s lan- guages according to structural properties such as basic word order (SVO, SOV, etc.) and the presence or absence of tone. On the one hand, typological features, by definition, allow us to compare an arbitrary pair of languages including language isolates. On the other hand, they pose several new challenges to us. While the sharing of cognates is a direct indicator of shared ancestry, the sharing of the same basic word order SOV, for example, is only a weak signal because it occurred multiple times in multiple places. We believe that computer-intensive statistical methods can help taming the inherent uncertainty. In this paper, we give a step forward toward typology-based phylogenetic inference. …

Content truncated to 500 words