Metonymy resolution (MR) is a challenging task in the field of natural language processing. The task of MR aims to identify the metonymic usage of a word that employs an entity name to refer to another target entity. Recent BERT-based methods yield state-of-the-art performances. However, they neither make full use of the entity information nor explicitly consider syntactic structure. In contrast, in this paper, we argue that the metonymic process should be completed in a collaborative manner, relying on both lexical semantics and syntactic structure (syntax). This paper proposes a novel approach to enhancing BERT-based MR models with hard and soft syntactic constraints by using different types of convolutional neural networks to model dependency parse trees. Experimental results on benchmark datasets (e.g., ReLocaR, SemEval 2007 and WiMCor) confirm that leveraging syntactic information into fine pre-trained language models benefits MR tasks.

We present two studies exploring how participants respond when a speaker contrasts two ideas, then expresses an ambiguous preference towards one of them. Study 1 showed that, when reading a speaker’s preference as text, participants tended to choose whatever was said last as matching the speaker’s preference, reflecting the recent-mention bias of anaphora resolution. In Study 2, we asked whether this pattern changed for audio versions of our stimuli. We found that it did not. We then asked whether observers used gesture to disambiguate the speaker’s preference. Participants watched videos in which two statements were spoken. Co-speech gestures were produced during each statement, in two different locations. Next, an ambiguous preference for one option was spoken. In ‘gesture disambiguating’ trials, this statement was accompanied by a gesture in the same spatial location as the gesture accompanying the first statement. In ‘gesture non-disambiguating’ trials, no third gesture occurred. Participants chose the first statement as matching the speaker’s preference more often for gesture disambiguating compared to non-disambiguating trials. Our findings add to the literature on resolution of ambiguous anaphoric reference involving concrete entities and discourse deixis, and we extend this literature to show that gestures indexing abstract ideas are also used during discourse comprehension.

Referential expressions (REs) have been investigated in L2 English but to date there is no single study that systematically and simultaneously analyzes the development and acquisition of the multiple factors that constrain the choice of REs in natural discourse production. We investigate L1 Spanish–L2 English learners across three proficiency levels versus an English control group from the COREFL corpus. An analysis of both the RE and its antecedent(s) reveals that different intra- and extralinguistic factors constrain the choice of REs (information status, activated antecedents, syntactic configurations, characterhood, within-task effect, and proficiency level). L2 learners (L2ers) are sensitive to some factors but are unable to fully attain native-like levels even at advanced stages. They do not transfer null subjects from their L1 contrary to previous L2 research, and do not find all contexts at the syntax-discourse interface equally problematic, thus confirming previous theoretical proposals and empirical findings.