Constructicography at work

Lyngfelt, Benjamin; Bäckström, Linnéa; Borin, Lars; Ehrlemark, Anna; Rydstedt, Rudolf

doi:10.1075/cal.22.03lyn

Cited by 19 publications

(6 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, the underlying databases of SAOL and SO have been migrated into the Karp lexical platform, which has been under active development for over a decade as a tool for working with formally structured language data [29], notably the computational lexical databases making up SweFN++ (in particular the Swedish FrameNet [30] and the Swedish constructicon [31]). The migration has also resulted in a long sought-after union, and to some extent harmonization, of the two sibling print-dictionary database structures (SAOL and SO).…”

Section: The Present: Innovata Traderementioning

confidence: 99%

Tradita innovare, innovata tradere

Borin,

Holmer

2024

Linköping Electronic Conference Proceedings

View full text Add to dashboard Cite

Swedish computational lexicography has a long history at the University of Gothenburg, both in its primary role as a central aspect of the scientific study of vocabulary and also as an infrastructural component for conducting research based on language data. Starting in the 1960s, the Språkdata research group pioneered corpus-supported lexicography for Swedish, forming the basis for successive editions of the two main descriptive dictionaries of contemporary Swedish, SAOL and SO. Language technological lexical resources for Swedish have been developed by the research unit/research infrastructure Språkbanken Text since the turn of the millennium, most recently in the framework of the Swedish FrameNet++ initiative. After two decades of separation, these two largely mutually independently developed strands of computational lexicography have now joined forces under the umbrella of Språkbanken’s lexical research infrastructure to advance the field technically, methodologically, and scientifically.

show abstract

Section: The Present: Innovata Traderementioning

confidence: 99%

Tradita innovare, innovata tradere

Borin,

Holmer

2024

Linköping Electronic Conference Proceedings

View full text Add to dashboard Cite

show abstract

“…Karp has an editing mode (Borin et al, 2013b), which was originally developed for building the Swedish FrameNet (Borin et al, 2010), but which has since been extended into a general editing environment for formally structured language data. Notably, the Swedish and Russian constructicons are built using the Karp editor (Lyngfelt et al, 2012;Janda et al, to appear 2018), as is the Swedish Women Online biographical database. 6 See Figure 4.…”

Section: Lsi Tables and Specimens As Lexiconsmentioning

confidence: 99%

Selected papers from the CLARIN Annual Conference 2022

Simov¹,

Eskevich²

2023

Linköping Electronic Conference Proceedings

View full text Add to dashboard Cite

i abstracts. Each submission was reviewed by at least three reviewers, all of whom scientists with rich experience and significant achievements in Language Technology and/or Digital Humanities. 24 submissions were accepted as a result of the two reviewing and correction rounds: 13 for oral presentations, the remaining ones as posters. The selected contributions were published online in the Book of Abstracts: https://www.clarin.eu/content/abstracts-overview-clarin-annual-conference-2017.After the conference, the authors of all accepted presentations were invited to prepare significantly expanded versions of their abstracts as full paper proposals. The authors were strongly encouraged to take into account the outcome of the discussion of their presentations during the conference, as well as their further work performed after conference. 12 papers were submitted and reviewed by the Program Committee. Each paper was reviewed by two reviewers, during two iterations. The process was completed by a final review with the aim to check if all the requested changes had been introduced. On the basis of the improved versions sent after the second reviewing iteration, all 12 papers were accepted and included in this volume.The accepted papers represent an overview of the most important topics discussed during the conference. The first two papers (Nicolas et al.) and (Parisse, et al.) report on the process of the formation of national consortia and their initial development.Next, Durco et al. revisit an important problem of constructing mapping between metadata formats from the point of view of building links between the CLARIN CMDI metadata standard and the metadata system in the PARTHENOS project. One of the goals of this project is to build a platform linking several major research infrastructures related to the area of SS&H. Metadata mapping is also close to a very challenging issue of the curation of metadata content.Zinn presents further development of the Switchboard systems for effortless and automated linking of resources with language tools, as well as processing chains on the basis of resources content. A new mechanism of directly linking EUDAT's B2DROP cloud service to Switchboard is the main focus of the paper.The next papers, namely of Sugimoto and Monachini et al., are about a problem of great importance for CLARIN (and every research infrastructure), i.e. studying and discovering users' needs. Sugimoto analyses users' behaviours through the observation of their activities in the already existing infrastructure and its systems. Monachini et al. report on an interesting surveybased study among the researchers in Classics Studies on the actual and intended use of digital methods. The results and conclusions are also illustrated by a mock-up prototype of a future system.The papers of Fišer & Lenardič and Borin et al. are devoted to the enrichment of language resources in CLARIN on the basis of the already existing resources in a way focused on the identified users' needs. Fišer & Lenardič report on the results of a...

show abstract

“…for more details on the tools used). Indeed, for the creation of the Swedish constructicon, Lyngfelt, Bäckström et al (2018) used computational tools to automatically extract constructions from large datasets. Their focus has mostly been on partially schematic constructions, that is, constructions that have one fixed lexical element but the surrounding slots are more schematic in that they can be filled by various lexical elements usually belonging to the same or a similar category, e.g.…”

Section: Identifying Constructions In Building Constructiconsmentioning

confidence: 99%

“…While BERTology has its own set of limiting factors that must be taken into account when interpreting these results (Rogers et al 2020), the fact that such models are already capable of making decisions about what counts as a construction is very promising, and there is no doubt that further collaboration between construction grammarians and computational linguists can only improve these models' ability to identify, retrieve, and classify constructions. A significant step in this direction is the work conducted by Lyngfelt, Bäckström et al (2018) in the automatic retrieval of strings/structures which they then checked manually.…”

Section: Discovering Constructional Candidatesmentioning

confidence: 99%

CxGBERT: BERT meets Construction Grammar

Madabushi

Romain²,

Divjak³

et al. 2020

Proceedings of the 28th International Conference on Computational Linguistics

View full text Add to dashboard Cite

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.

show abstract

Constructicography at work

Cited by 19 publications

References 34 publications

Tradita innovare, innovata tradere

Tradita innovare, innovata tradere

Selected papers from the CLARIN Annual Conference 2022

CxGBERT: BERT meets Construction Grammar

Contact Info

Product

Resources

About