The background and key concepts are presented, including the notion of salience of information and an overview of RRG as a parallel architecture theory, in particular the RRG representations of discourse-pragmatics as a separate and independent component. The important feature of RRG is the bidirectional interactions between the separate components of grammar, semantics, syntax, and discourse, and these interactions are discussed with examples. The study applies the notion of salience to the RRG theory of the syntax–semantics–discourse interface, and, for this, utilizes the bidimensional model of salience, consisting of both backward-looking and forward-looking salience. This bidimensional characterization is important because it captures both the hearer-based and the speaker-based discourse models. Furthermore, the study introduces the notion of forward-looking non-salience, the speaker’s active backgrounding of information, which is essential to the account of zero-marking and postposing of arguments in Japanese.

While summarizing the main points, the concluding chapter revisits the bidimensional model of salience based on the study of Japanese discourse and its implications for RRG, which is followed by remarks on the findings in the study of L2 Japanese discourse. Overall, zero and topic coding of arguments and zero coding of predicates mostly pertain to backward-looking salience, while predicate-less repetitions in L1 discourse relate to the forward-looking dimension of salience. On the other hand, zero marking and post-predicate coding of arguments pertain to forward-looking salience, specifically to imposing non-salience. With these, the study provides abundant evidence for the RRG claim that discourse-pragmatics plays a role in virtually every part of grammatical systems. Furthermore, the analysis of both backward-looking and forward-looking salience demonstrates the bidirectional discourse–syntax interfaces, an important property of the parallel architecture theory, because the backward-looking and forward-looking dimensions are relevant to the discourse-to-syntax and syntax-to-discourse linking respectively.