The aims of this study are to investigate several discourse functions of nɪ-marked expressions used in narrative discourse in Abou Dida (Eastern Kru, Côte d’Ivoire) and to explore the possibility of expanding the research horizons of cognitive linguistic views of discourse phenomena building on the observations of how nɪ-marked expressions behave in the unfolding of discourse. While the discourse functions of nɪ-marked expressions can fall under the classification of left-dislocation, they are not exactly the same as what has been discussed in the preceding studies; left-dislocation has not been a central issue in cognitive linguistics. What makes nɪ-marked expressions unique is that nɪ can both left-dislocate and topicalize lexical NPs, forming a continuous spectrum of discourse functions. What cognitive linguistics contributes to left-dislocation studies is that it can generalize functions and patterns beyond individual examples. The survey results tell us about the significance of an empirical discourse analysis in a generalizable theoretical setting.

As an attempt at a data-oriented cognitive-typological study of language, we present an analysis of the linguistic representations of changes of state in Japanese. The expressions of 12 different changes of state found in BCCWJ are analyzed in a framework also used for the typology of motion-event descriptions (Matsumoto 2017a; see also Talmy 2000). The study reveals that a change of state (or at least the transition of change) is expressed overwhelmingly in the head position of a sentence; very often the transition and the resultant state are expressed together by the main verb. In this respect, Japanese can be said to be a “head change-coding language.” Compared with the expressions of spatial motion, which similarly use the head position to encode path, expressions of state changes encode change very often in the head only, testifying to a greater role of the head for changes of state. There are large variations in the degree to which the head is used, depending on the type of state change, inviting an explanation of what sort of state and change of state are expressed with the use of verbs as opposed to adjectives. Furthermore, the low frequency of “co-events” occurring with changes of state demonstrates the limitations of Talmy’s typology.

The present paper investigates how the so-called better off construction, an irregular idiomatic construction, emerged in the history of English from three perspectives: cognitive linguistics, corpus linguistics, and natural language processing (NLP). It then demonstrates various empirical methods to discuss constructions using deep learning. To that end, this paper (i) conducts extensive corpus research to examine the emergence of the better off construction, (ii) analyzes the data from a cognitive linguistic perspective to develop a hypothesis, and (iii) tests the legitimacy of the hypothesis using linguistic and NLP methods such as word vectors. Additionally, through the research on the better off construction, this paper attempts to (1) present a ‘research cycle’ for empirical linguistic analysis by applying various methods from NLP, (2) show the similarities between usage-based analysis and deep learning, and (3) investigate the possibility that deep learning can provide converging evidence for interdisciplinary linguistic analysis.

Instrument verb formation is very productive in English (e.g., knife_V the man), whereas it is restricted in some way in Japanese (e.g., *naihu-(su)ru ‘to knife’). From the perspective of Frame Semantics (Fillmore 1982, among others), this study observes that central to the formation of instrument verbs is to associate a certain scene with a thing named by a parent noun (i.e., a noun to be converted into a verb), arguing that this difference in productivity between English and Japanese reflects their unique manner of linguistic encoding which is motivated by their basic egocentricity (Hirose 2000). More precisely, in Japanese as a private-self-centered language, encoders must overtly encode how the parent-noun participant is associated with (or used in) a scene (e.g., naihu-de-kiru/sasu/osô ‘to cut/stab/attack with a knife’), which corresponds to the “off-stage” (Boas 2003) information with respect to the scene. By contrast, in English as a public-self-centered language, decoders can view the associated scene on a par with the encoder and thus activate off-stage information as to how and for what purpose the parent-noun participant is used within the scene, even without overt linguistic encoding of the relevant information. As a result, Japanese “exceptional” instrument verbs (e.g., hôtyô-suru, hottikisu-suru), even if occurring seemingly at random, are classified in a principled way as follows: (i) register-specific instrument verbs that presuppose who is talking to whom and (ii) what Hirose (inter alia 2000) calls “private expressions” that may be employed as a strategy to express intimacy to the addressee.