The FrameNet team gave a tutorial entitled "Frame Semantics across Languages: Towards a Multilingual FrameNet" at the International Conference on Computational Linguistics (COLING 2018) in Santa Fe, New-Mexico on the morning of August 20th. Attendees included researchers ranging from newcomers to FrameNet to those who have been developing software using FrameNet data for years, and they offered many useful ideas and suggestions to each other and to the presenters.
Final versions of the slides are now available on the link above. Also, an updated bibliography for the entire tutorial (including some resources not mentioned in the slides) is available here (PDF)
The FrameNet Project at the International Computer Science Institute (http://framenet.icsi.berkeley.edu) has built a lexical database for contemporary English, defining over 1,220 semantic frames containing more than 13,600 lexical units (word senses), whose valence descriptions are documented by more than 202,000 manually annotated examples. Natural Language Processing researchers and developers have shown that Frame Semantic representations are useful in applications such as question answering, text-to-scene systems, spoken dialog systems, and social network extraction.
Separate research projects have developed Frame Semantic lexical databases for approximately a dozen more languages (including Brazilian Portuguese, Chinese, Hebrew, Korean, and Swedish) based in varying degrees on the original FrameNet database structure (Baker et al. 2003), frame descriptions, etc., and on FrameNet methodology and annotation practices (Ruppenhofer et al. 2016).
The goals of the Multilingual FrameNet Project (NSF Award #1629989) are: (1) to align the databases of the different FrameNets at the level of frames and that of words (technically, lexical units, i.e., a single sense of a word in a frame); and (2) to investigate the extent to which the semantic frames of the English FrameNet database might be construed as cross-culturally and cross-linguistically “the same” and to determine the underlying linguistic and cultural differences when they differ. Multilingual FrameNet has also begun a parallel annotation task, which requires addressing a range of questions involving representation, cross-lingual framing differences, metonymy, and metaphor, as well as the interaction of frames and constructions (Fillmore 1988, 2013).
Participants in this tutorial learned about:
- FrameNet, its methodology, and practices
- Cross-linguistic similarities and differences among the languages
- Principles of multilingual alignment and the challenges involved
- Issues of Representation in frame vs. constructional analyses
- Potential Applications of a multilingual FrameNet database
Outline of Tutorial
- Frame Semantics and FrameNet (Miriam R. L. Petruck)
- Frames, Frame Elements (roles), and Lexical units
- Annotation and Reports
- Frame Relations and Frame Element Relations
- Construction Grammar and Constructicons
- The Multilingual FrameNet Project (Collin Baker)
- FrameNets in Languages other than English
- Variations in Goals and Practices
- Alignment across Languages
- Alignment Algorithms
- Parallel Annotation on TED Talk (Michael Ellsworth)
- Web Annotation Tool (Torrent et al. 2016)
- Cross-lingual Framing Differences
- Constructions implicit in FrameNet Annotation
- Trade-offs in Representation
- Metaphor and Metonymy in FrameNet
- Mental Spaces: Negation and Conditionals
- Uses of Multilingual FrameNet (Swabha Swayamdipta)
- Automatic Semantic Role Labeling
- Automatic Semantic Role Labeling (English)
- Automatic Semantic Role Labeling (Other Languages)
- Multilingual Applications
- Automatic Semantic Role Labeling
- Representing Semantics in Frames and Constructions
- Frame-based Knowledge Representation
- Interactions between Frames and Constructions
- Implications for Multilingual Applications
Collin F. Baker
Collin F. Baker (International Computer Science Institute, email@example.com, https://www.icsi.berkeley.edu/icsi/people/collinb) has been affiliated with the FrameNet Project since it began (as Project Manager, 2000-present), working closely with the late Charles J. Fillmore and many collaborators. He was also Project Manger of the MetaNet Project at ICSI (2012-2015, http://metanet.icsi.berkeley.edu), an IARPA-funded effort to recognize metaphoric language automatically in several languages. His research interests include building FrameNets in other languages (Loenneker & Baker2009), aligning FrameNet to other lexical resources (Fellbaum & Baker 2013,Ferrandez et al. 2010, Fellbaum & Baker 2008), and linking to ontologies and reasoning (Scheffczyk et al. 2010).
Michael J. Ellsworth
Michael Ellsworth (International Computer Science Institute, (firstname.lastname@example.org, https://berkeley.academia.edu/MichaelEllsworth) has been involved in lexical semantic research for nearly 20 years on the FrameNet Project, starting with a summer internship in the original FrameNet project. In this time he has become a key member of the FrameNet team, having been involved in the process of frame definition, annotation, annotator training, and data-integrity checking. Since 2002, he has been the researcher in charge of the ontology-like hierarchy that organizes the frames. Publication topics include the differences between FrameNet and other annotation projects, the FrameNet hierarchy and ontologies, the principles behind FrameNet annotation, paraphrasing using FrameNet, and various English grammatical constructions. He is currently writing a dissertation on how the domain of emotion is encoded in English words and grammatical constructions.
Miriam R. L. Petruck
Miriam R. L. Petruck (International Computer Science Institute, email@example.com, https://www.icsi.berkeley.edu/icsi/people/miriamp) received her Ph.D. in Linguistics from the University of California, Berkeley, CA, under the direction of the late Charles J. Fillmore. A key member of the team developing FrameNet almost since the project’s founding (in 1997), her research interests include semantics, lexical semantics, knowledge base development, grammar and lexis, semantics, Frame Semantics and Construction Grammar, particularly as these linguistic theories support advances in NLU and NLP. She is a frequent invited speaker, lecturing internationally about Frame Semantics, Construction Grammar, and FrameNet.
Swabha Swayamdipta (firstname.lastname@example.org, http://www.cs.cmu.edu/~sswayamd) is a PhD student at the Language Technologies Institute at Carnegie Mellon University. She works with Noah Smith and Chris Dyer on developing efficient algorithms for broad-coverage semantic parsing, with a focus on exploiting the relationship between syntax and semantics. Her research interests also include applications of broad-coverage semantics for tasks such as entailment and coreference. She has a Masters degree from Columbia University, and was a former research intern at Google New York. She is currently a visiting student at University of Washington in Seattle.