Similarity Semantics and Building Probabilistic Semantic Maps from Parallel Texts

Abstract

This paper deals with statistical (non-implicational) semantic maps, built automatically using classical multidimensional scaling from a direct comparison of parallel text data (the Gospel according to Mark) in the domain of motion events (case/adpositions) in 153 languages from all continents in 190 parallel clauses. The practical objective is to present one way (among other possible ways) in which semantic maps can be built easily and fully automatically from large typological datasets (Section 3). Its methodological objective is to demonstrate that semantic maps can be built in various ways and that the sampling of languages and small differences in the method chosen to build a semantic map can have a strong influence on the results (Section 4). This does not mean that semantic space is arbitrary, but rather that it is dynamic (having stretching and shrinking dimensions). The theoretical aim of this paper is to discuss similarity semantics, the implicit theoretical basis behind the semantic map approach, and to show that similarity semantics is not novel, but has a long-standing tradition in philosophy and psychology (Section 2).

Predicative reduplication: Functions, their relationships and iconicities

“…Where large numbers of constructions are involved, other methods for representing semantic affinity may be used (Cysouw 2007;Croft & Poole 2008;Wälchli 2010;Regier et al 2013;Levshina 2016).…”

Proceedings of the 6th Joint Conference on Lexical and Computational Semantics (*SEM 2017)

“…The Preposition Project (TPP; Litkowski and Hargraves, 2005) broke ground in stimulating computational work on fine-grained word sense disambiguation of English prepositions (Litkowski and Hargraves, 2005;Ye and Baldwin, 2007;Tratz and Hovy, 2009;Dahlmeier et al, 2009). Typologists, meanwhile, have developed semantic maps of functions, where the nearness of two functions reflects their tendency to fall under the same adposition or case marker in many languages (Haspelmath, 2003;Wälchli, 2010).…”

Mapping out the Source domain

“…To achieve our goals we exploit the availability of parallel, aligned corpus data for our four languages and generate the maps without semantic preconditioning by way of multidimensional scaling (MDS), along the lines of Wälchli (2010). We find that the maps give us a clear tripartite division of the semantic space into Source, Location, and Goal, that the languages have fairly different and clearly independent means of expressing Source, and that there are numerous interactions and overlaps between Source and Location, Goal and also various non-spatial semantic domains, respectively.…”