Introduction

When the history of historical linguistics in the twentieth century is written, one recurrent theme will surely be the hypothesis that certain types of change are unidirectional. This hypothesis takes many forms, but is probably most widely associated with historical cross-linguistic, typological work, much of it devoted to the correlations among changes in meaning and morphosyntax known as grammaticalisation (see e.g. Greenberg 1978; Lehmann 1995 [1982]; Hopper and Traugott 1993; Bybee, Perkins and Pagliuca 1994). Critics of the hypothesis have pointed out that unidirectionality is not exceptionless and can be reversed (Joseph and Janda 1988; Newmeyer 1998; Lass 2000, among others). Being social as well as cognitive, and subject to contingencies such as production, perception, transmission and social evaluation, no change is likely to be exceptionless. Unidirectionality is a strong tendency manifested by particular sets of changes. The present study is a further contribution to the debate on unidirectionality, with focus on evidence for it in semantic change. The hypothesis is that intersubjectification, in the sense of the development of meanings that encode speaker/writers’ attention to the cognitive stances and social identities of addressees, arises out of and depends crucially on subjectification. Schematically, subjectification > intersubjectification, not intersubjectification> subjectification. This is a semasiological hypothesis about constraints on the kind of changes that individual lexemes may undergo. It also has implications for onomasiological constraints on shifts of meaning from one conceptual domain to another, e.g. from the domain of spatial position to politeness marker in Japanese, but not vice versa.