Toward a Redefinition of Yes/No Questions

While both theoretical and empirical studies of question-answering have revealed the inadequacy of traditional definitions of Ve*-no questions (YNQs), little progress has been made toward a more satisfactory redefinition. This paper reviews the limitations of several proposed revisions. It proposes a new definition of YNQs baaed upon research on a type of conversational irnpIieature, termed here setdar imp/ie,,ture, that helps define appropriate responses to YNQs. By representing YNQs as sealer qtteriee it is possible to support a wider variety of system anti user responses in a principled way. I INTRODUCTION If natural language interfaces to question-answering systems are to support a broad range of responses to user queries, the way these systems represent queries for response retrieval should be reexamined. Theorists of question-answering commonly define questions in terms of the set of all their possible [true) answers. Traditionally, they have defined t/us-no quesgiofts (YNQs) as propositional questions (?P) or as a special type of alternative question (?P V ?Q), in which the second alternative is simply the negation of the fir.~t (?P V ?"P). So 'Does Mary like skiing?' would be represented as flikr(lffary,skling) or ?like(Alary.okiing) V "~-Iikt(Mary, skiing) and the range of appropriate responses wouhl be gee, no and, possibly, unknown. • " ilowever, both theoretleal wnrk and empirical studies of naturally occurring question-answer exchanges have shown this approach to be inadequate: ?~s, .o. and unknown form only a small portion of the set ¢,f all appropriate responses to a YNQ. Furthermore, for some YNQ's. none of these simple direct responses alone i~ appropri:,te. While it is widely recognized (llobbs, 1979, Pollack, 1982} that indirect resp.nses I to YNQs represent an important option for respondents in natural discourse, standard theories of question-answering have n,~t been revised accordiugly. A practical COllsPquence surface~ when attempts are made t.o support indirect responses to YNQs computationally. For lack of alternative representations, question-answering systems which would permit indirect responses must still represent YNQs as if the direct respons-s were the 'norm', and then resort to ad hoe manipulations to generate second-class 'indirect' responses, thus perpetuating an asymmetric distinction between 'direct' and 'indirect' responses. However. resea.'h under way on how a type of generalized conversational implieatttre, termed here scalar irrtplieature, can be used to guide the generation and iaterpretion of indirect respt,nses to YNQs sugges/,s a revi~ed representation fi)r YNQs which scrotums)dates a wide variety of responses in a uniform way. is one of the …