Where does gender come from? Evidence from a complex inflectional system

Abstract

Although inflectional morphology has been the focus of considerable debate in recent years, most research has focused on English, which has a much simpler inflectional system than in many other languages. We have been studying Serbian, which has a complex inflectional system that encodes gender, number, and case. The present study investigated the representation of gender. In standard theories of language production, gender is treated as an abstract syntactic feature segregated from semantic and phonological factors. However, we describe corpus analyses and computational models which indicate that gender is correlated with semantic and phonological information, consistent with other cross-linguistic studies. The research supports the idea that gender representations emerge in the course of learning to map from an intended message to a phonological representation. Implications for models of speech production are discussed.

Acknowledgments

This research was supported by NIMH grant RO1 MH58723 and NIMH grant P50 MH64445. Mark Seidenberg was also supported by an NIMH research scientist development award. We would like to thank Marc Joanisse with whom we collaborated on developing the models used here, using software developed by Michael Harm, whom we also thank. We also thank Aleksandar Kostić for making available the electronic version of the Frequency Dictionary of Contemporary Serbian. This work benefited from the feedback of the attendees of the 3rd Workshop on Morphological Processing, held in Aix-en-Provence, France.

Notes

¹Languages referred to as Serbian, Croatian, and Bosnian were referred to as Serbo-Croatian in earlier literature.

²All Serbian examples are written in International Phonetic Alphabet, unless otherwise specified.

³This larger corpus consists of 2 million words from the 20th century daily press and poetry (Kostic, 1999, and http://www.serbian-corpus.edu.yu).