Handling compound nouns in a Swedish speech-understanding system

Abstract

This paper describes and evaluates a simple and general solution to the handling of compound nouns in Swedish and other languages in which compounds can be formed by concatenation of single words. The basic idea is to split compounds into their components and treat these components as recognition units equivalent to other words in the language model. By using a principled grammar-based language-processing architecture, it is then possible to accommodate input in split-compound format. 1

Dutch speech recognition in multimedia information retrieval

“…The comparison of systems that apply compound splitting with systems that do not, deserves special attention. As noted in Carter et al (1996), WER of systems applying compound splitting can be measured in two ways: 1. by taking the speech recognition hypothesis after compound splitting and comparing this with a reference in which compounds are split as well (split comparison), 2. by mapping the compound constituents in the speech recognition hypothesis after compound splitting back to the original compounds, and comparing this with the original reference (unsplit comparison).…”

“…It was however included as it can be argued that with this method split systems and unsplit systems can better be compared from a language processing point of view. This method was also regarded to be of primary relevance in the study of Carter et al (1996). Table 10.5: OOV rates and WER rates of an unsplit system and split systems: unrestricted refers to a compound splitting procedure that splits all encountered compounds, restricted refers to procedures that split compounds only when they do not occur in the top N words of a sorted word frequency list.…”