Bias, elicitation and endangered language description

While many have been focussed on methodological issues of documenting endangered languages, others have asked ‘How will this current focus on documenting endangered languages influence language description, and language typology?’ In this paper I will discuss how language endangerment may lead to biases in the data linguists collect for analytical and descriptive purposes, and the consequences of this for language description and language typology. Elicitation tools offer insight into language paradigms which may be difficult to document when relying on purely naturalistic data. They have a long history of use in descriptive linguistic fieldwork. Sometimes the use of such tools is the only way to collect data about specific domains and paradigms of severely endangered languages, where there is minimal opportunity to observe naturalistic language use. The endangered language context may also present a different set of challenges when it comes to using guided elicitations. Some elicitations require ‘controlled experiment’ conditions, which are unfamiliar and/or uncomfortable for elderly speakers. In such situations, there may be a preference to work with (younger) speakers who are more comfortable with elicitation tasks. The end result of this will be a bias in the data set towards their data, or their variety of the language. In this paper, I will discuss my own experiences with ‘bias’ when using the MPI demonstrative questionnaire (Wilkins 1999), with Dalabon speakers. Dalabon (Gunwinyguan, Australia) is a severely endangered language of south-western Arnhem Land. I will argue for the value of these tools for the insight they offer into linguistic systems, and discuss the resultant issues for analysis, description and typology which arose from relying heavily on data from younger semi-speakers. Working with younger semi-speakers produced a hypothesis I was able to test against data from the older fluent speakers, but at the same time, increased the possibility of recording language features which are inauthentic, or the product of interference from another code. For example, the younger semi-speaker's data contained a particular syntactic construction not seen previously in the data from older speakers. Bias is inherent in the collection of linguistic data, and I argue that the linguist must exercise extra care when using elicitations when documenting endangered languages. Wilkins, David P. 1999. The 1999 Demonstrative Questionnaire: "THIS" and "THAT" in comparative perspective. In "Manual" for the 1999 field season, edited by Stephen C. Levinson and N. J. Enfield. Nijmegen: Language and Cognition Group, Max Planck Institute for Psycholinguistics