LFG parse disambiguation for Wolof

Friday February 14, 2014 at 14:15 – 16:00 (HF: 217)

In this presentation, I will discuss ambiguity issues arisen during the design and implementation of a computational large-scale LFG grammar for Wolof (Niger-Congo language mainly spoken in Senegal, ca. 10 millions speakers). The LFG grammar has been developed as part of my PhD project which deals with building language resources and tools for Wolof. The grammar is implemented with XLE and tested on natural language data.

In the course of grammar development, many ambiguity issues arose via alternative definitions of morphological and lexical entries, and from syntactic ambiguities. Thus, the discussion will focus on three kinds of ambiguity: 1) morphological, 2) lexical and 3) syntactic ambiguities. I will show how Wolof nouns constitute a typical source of ambiguity due to polysemy and homonymy. I will argue that Wolof nouns exhibit noun class underspecification to be considered as a case of feature indeterminacy (Dalrymple, 2009). Following on from this, I will provide evidence that the noun class structures show similarities with the CASE underspecification phenomena as has been observed for German (Dalrymple, 2009).

Concerning issues related to lexical and syntactic ambiguities, I will discuss disambiguation methods based on Constraint Grammar and the c-structure pruning algorithm of XLE as possible models to increase parsing efficiency and performance. I will describe experiments conducted on Wolof and discuss the benefits of applying such methods to control/reduce ambiguity in the LFG grammar.

By Cheikh Bamba Dione, PhD Candidate, Research Group: Language Models and Resources, LLE