Urdu/Hindi Motion Verbs and Their Implementation in a Lexical Resource

Abstract

A central task of natural language processing is to find a way of answering the question "Who did what to whom, how, when and where?" with automatic means. This requires insights on how a language realizes events and the participants that partake in them and how this information can be encoded in a human- as well as machine-readable way. In this thesis, I investigate the ways that the spatial notions of figure, ground, path and manner of motion are realized in Urdu/Hindi and I implement these insights in a computationally-usable lexical resource, namely Urdu/Hindi VerbNet. I show that in particular the encoding of complex predicates can serve as a guiding principle for the encoding of similar constructions in other VerbNets.This enterprise involves a detailed investigation of the syntax-semantics interface of motion verb constructions in Urdu/Hindi, in particular the different syntactic alternation patterns that realize motion events. As it turns out, Urdu/Hindi employs complex predicates of motion that denote the manner of motion along a path with two verbal heads. This construction exhibits similar syntactic properties as aspectual complex predicates in the language (Butt 1995). The thesis shows that the combinatorial possibilities between main verb and light verb are driven by the manner/result complementarity established by Levin and Rappaport Hovav (2008, 2013), according to which verbs either lexicalize non-scalar manner of motion or denote a scalar result event. An analysis of the construction in Lexical-Functional Grammar (Bresnan and Kaplan 1982, Dalrymple 2001) shows that the two predicates merge their arguments at the level of argument structure, which in turn can be mapped onto the functional representation along the lines of Bresnan and Zaenen (1990). From a typological point of view, the combination of two verbal heads denoting manner of motion along a path in a monoclausal construction shows that Urdu/Hindi belongs to the group of equipollently-framed languages (Slobin 2004, 2005).A key component of the syntax-semantics interface, in particular with respect to the combination of syntactic and semantic information in Urdu/Hindi VerbNet, is the notion of event structure. Here, I differ with respect to other VerbNets in following an approach which assumes that syntax has predictable meaning correlates. In particular, I analyze Urdu/Hindi motion verbs in the event-decompositional frameworkof the first phase (Ramchand 2008) and show that the approach provides the right level of abstraction for encoding event-structural information on Urdu/Hindi verbs in a computational lexical resource. It makes the right design decisions as to the event structure of the different syntactic alternations that Urdu/Hindi motion verbs can appear in. In particular, it accounts for complex predicates of motion and allows for a consistent theoretical-linguistic treatment of the phenomena, which provides an optimal basis for encoding the construction in the lexical resource.Urdu/Hindi VerbNet follows the general assumption of other VerbNets, e.g. English VerbNet (Kipper-Schuler 2005, Kipper et al. 2008) and Arabic VerbNet (Mousser 2011), in that semantically similar events are identified by a common syntactic structure. However, due to the multitude of syntactic patterns available in the class of motion verbs in Urdu/Hindi, this principle cannot be ported straightforwardly to Urdu/Hindi. In order to overcome this issue, I combine the theoretical linguistic insights gained in this thesis with those of a computational clustering approach: In particular, the subclasses arising out of the manner/result complementarity are further subclassified by employing an algorithmic visualization system that automatically clusters verbs according to their syntactic and semantic properties. The system offers an interactive way of investigating the resulting clusters and their coherency. This grouping of verbs is the basis for the implementation of the class of motion verbs in Urdu/Hindi VerbNet.In conclusion, the factors that determine the syntax-semantics interface of Urdu/ Hindi motion verbs allow for a multitude of theoretical insights on motion verb expressions in the language. These insights are used in encoding the concept of motion in Urdu/Hindi VerbNet, a resource that explicitly encodes the relation between syntactic and semantic information and complements other computational lexical resources for the language. The theoretical as well as computational linguistic analyses of the syntax-semantics interface of Urdu/Hindi motion verbs yield insights into the concrete realization of spatial notions in the language, namely the ways of encoding the figure, the ground, the path and the manner of motion; concepts that are present in motion verb expressions across languages.