Several important findings emerged from the cross-linguistic comparison of speech reporting strategies used in narrative discourse.
First, our exploration of the phenomenon of logophoricity in its natural discourse environment helped us to identify a number of problems with previous characterizations of logophoric pronouns, particularly challenging the notion of the logophoric pronoun as an element of indirect discourse and the assumption that constructions with logophoric pronouns share properties with both direct and indirect speech. Our data allowed us to show that African-style logophoricity differs in important ways from phenomena related to logophoric reference in European languages, leading us to believe that African-style logophoricity is a lexical rather than a syntactic phenomenon, and is best described as part of the language-specific system of personal deixis.
Second, our data suggest that the distinction between direct and indirect speech, which is central to discourse reporting in European languages, is not universal. The principles underlying this distinction in European languages differ from those underlying the choice between reported speech constructions in other languages. In particular, the traditional notion of indirect discourse is, on closer inspection, a complex one, involving a rather peculiar combination of parameters relating to different levels of linguistic structure (such as syntactic subordination and special use of deictic elements); the different parameters often operate independently in non-European languages. This observation challenges the “continuum” view of reported speech, which assumes that across languages reported speech constructions are organized along a continuum between two extremes, conveniently instantiated in European languages by “direct” and “indirect” speech.
Finally, our annotation method allowed us to perform a systematic statistical analysis of the distributions of words and constructions within and outside of reported speech, and to establish that word classes are unevenly distributed between reported speech and other parts of the narrative. Interjections, for example, are particularly likely to appear in reported speech, while ideophones only rarely do so. Asymmetric distributions allow us to identify subtle semantic distinctions between word classes and grammatical categories. Thus, interjections may be used by narrators to signal the beginning of reported discourse in much the same way as grammatical markers such as specialized quotative markers.
