The Problem of Polysynthesis in UMR Annotations: Complexities in Handling Preverbal Modification and Noun Incorporation in Arapaho

ABSTRACT: UMR (Uniform Meaning Representation) is a powerful new tool for representing the lexis-semantics interface graphically in a cross-linguistically uniform way. Its primary purpose is to support semantic parsing in natural language processing (NLP), but it is also a useful tool for documenting and analyzing semantic structures in language. In this paper, we demonstrate strengths and weaknesses of UMR in annotation of Arapaho, a highly polysynthetic and agglutinating Algonquian language. We describe lexicographical design tensions that emerge when applying UMR to languages like Arapaho and offer suggestions for refining UMR so that its goal of applying cross-linguistically can be fully realized. These tensions center on the need for a special Arapaho-UMR lexicon to guide annotators in carving up morphologically complex words into distinct predicate and argument structures in UMR graphs. Such a resource is critical for polysynthetic and agglutinating languages in which a single word may encode an entire complex proposition. We propose such a lexicon and demonstrate how it will help annotators stay as they navigate morphological complexity inside and outside the verb stem.