While lexico-semantic elements no doubt capture a large amount of linguistic information, it has been argued that they do not capture all information contained in text. This assumption is central to constructionist approaches to language which argue that language consists of constructions, learned pairings of a form and a function or meaning that are either frequent or have a meaning that cannot be predicted from its component parts. BERT's training objectives give it access to a tremendous amount of lexico-semantic information, and while BERTology has shown that BERT captures certain important linguistic dimensions, there have been no studies exploring the extent to which BERT might have access to constructional information. In this work we design several probes and conduct extensive experiments to answer this question. Our results allow us to conclude that BERT does indeed have access to a significant amount of information, much of which linguists typically call constructional information. The impact of this observation is potentially far-reaching as it provides insights into what deep learning methods learn from text, while also showing that information contained in constructions is redundantly encoded in lexicosemantics.

No abstract

This paper shows that low-level generalisations in argument structure constructions are crucial to understanding the concept of alternation: low-level generalisations inform and constrain more schematic generalisations and thus constructional meaning. On the basis of an analysis of the causative alternation in English, and more specifically of the theme (i.e., the entity undergoing the event denoted by the verb), I show that each construction has its own schematic meaning. This analysis is conducted on a dataset composed of 11,554 instances of the intransitive non-causative construction and the transitive causative construction. The identification of lower-level generalisations feeds into the idea that language acquisition is organic and abstractions are formed only gradually (if at all) from exposure to input. So far, most of the literature on argument structure constructions has focused on the verb itself, and thus fails to capture these generalisations. I make up for this deficit through an in-depth analysis of the causative alternation.

Full-fledged grammatical article systems as attested in Germanic and Romance languages are rather uncommon from a typological perspective. The frequency with which articles occur in these languages, together with the difficulty encountered in detecting them and the lack of a water-tight account of article use, make article errors one of the most frequent errors in language produced by L2 learners whose L1 does not feature an article system of similar complexity, all the while appearing unproblematic for L1 users. We present a conceptually and methodologically interdisciplinary approach to the grammatical category of articles in English and combine a usage-based, cognitive linguistic account of the function and use of articles that respects its discourse-based nature with a computational exploration of the challenges the system poses from the perspective of learning. Running a statistical classifier on a large sample of spoken and written discourse chunks extracted from the BNC and annotated for the five main determinants of article use reveals that Hearer Knowledge is the driver of a hierarchical system. Once Hearer Knowledge is acknowledged as the motivating principle of the category, article use becomes eminently predictable and restrictions are in line with the forms from which the articles have developed historically, with the and a acting as category defaults and zero acting as default override. Simulations with a computational model anchored in the psychology of learning shed light on whether and how human cognition would handle the proposed relations detected in the data. We find that different articles have different learnability profiles that, again, are in line with their historical development: while the can be learned from one strong indicator, the relationships for the zero article are less exclusive. On the basis of these findings, we argue that the article category appears as a referent tracking system that grammaticalizes the principles of “audience design”: it forces a speaker to track and mark reference from the vantage point of the memory of the hearer, thereby reducing the processing effort required from the hearer. This particular mindset inverses the typologically dominant situation in which this information is not explicitly marked by the speaker but implicitly retrieved from context by the hearer.

We examined how language supports the expression of temporality within sentence boundaries in English, which has a rich inventory of grammatical means to express temporality. Using a computational model that mimics how humans learn from exposure we explored what the use of different tense and aspect (TA) combinations reveals about the interaction between our experience of time and the cognitive demands that talking about time puts on the language user. Our model was trained on n-grams extracted from the BNC to select the TA combination that fits the context best. It revealed the existence of two different sub-systems within the set of TA combinations, a “simplex” one that is supported lexically and is easy to learn, and a “complex” one that is supported contextually and is hard to learn. The finding that some TA combinations are essentially lexical in nature necessitates a rethink of tense and aspect as grammatical categories that form the axes of the temporal system. We argue that the system of temporal reference may be more fruitfully thought of as the result of learning a system that is steeped in experience and organised along a number of functional principles.

This paper presents a method for quantitative and qualitative analyses of the causative alternation in English, where verbs may alternate between a transitive (causative) construction (Santa crinkled his eyes) and an intransitive (non-causative) construction (His eyes crinkled).1 The aim of this paper is to present a method designed to measure the alternation strength of causative verbs, i.e. the extent to which they alternate between the two constructions. One of the central elements this paper investigates is the Theme, i.e. the participant that is in subject position in the intransitive construction and object position in the transitive construction. A distinctive collostructional analysis (Gries and Stefanowitsch 2004) shows that certain verbs are significantly attracted to one of either two constructions while others are equivalently distributed in the two constructions. However, after careful analysis it appears that very few Themes actually overlap between the two constructions (Lemmens forthcoming) which indicates that each construction seems to be rather restrictive regarding which Themes they recruit. The low degree of alternation of the Themes leads us to ask ourselves the extent to which the alternation is part of a speaker’s knowledge of their language.