Dr. Rezvan Motavallian is an associate professor of Linguistics at the University of Isfahan. Her research interest includes formal syntax, syntactic structure and its interface with semantics, error analysis, and learner corpus.

Seminar: Applying of English Frames to Persian: Challenge and Solutions.
Date: April 4th, 2023.
Abstract: In the project presented, we intend to take a step towards designing a Persian FrameNet and examine the transferability of the parts of English FrameNet to Persian FrameNet. To accomplish this task, we consider the concept of motion due to its significance in cross-linguistic studies and its different lexicalization patterns in typologically diverse languages. In this regard, we investigate the challenges that different features of the Persian language and particularly the difference in the way of expressing motion events in this language makes for the compatibility of English FrameNet with Persian FrameNet. Focusing on Persian motion verbs, we try to determine the extent to which it is possible to transfer different components of the English FrameNet – including semantic frames, coreness and peripherality status of Frame Elements, syntactic and semantic structures and frame-to-frame relations to other languages. The results suggest that English FrameNet is prone to incompatibility due to the lack of systematicity in the specification of frames and frame-to-frame relations. Moreover, different polysemy processes, the presence of the same word as different lexical units, various syntactic alternations of lexical units, different levels of importance of Frame Elements in lexicalization in various languages,  the tendency of some languages like Persian to combination and incorporation,  the presence of a large number of compound verbs as well as the impact of culture make challenges for the transferability of English FrameNet to other languages.

Dr. Svetla Koeva is a professor of Computational Linguistics and the head of the Department of Computational Linguistics at the Institute for Bulgarian Language, Bulgarian Academy of Sciences. Her research interests lie in the field of computational linguistics, formal description of language: morphology and syntax, lexical-semantic networks and ontologies. She is the principal investigator in the development of various language resources for Bulgarian such as: Bulgarian WordNet, Bulgarian National Corpus, and Bulgarian FrameNet.

Seminar: Bulgarian FrameNet: current state and future prospects.
Date: October 4th, 2022
Abstract: The talk covers: (a) identification of Bulgarian verbs that evoke a particular FrameNet semantic frame; (b) language specific data encoded to describe the Bulgarian verbs (mainly, values of grammatical categories and grammatical roles); c) mapping with the FrameNet semantic frames: equivalence (occurs when the lexical units in the two languages are translation equivalents and the semantic frame accurately represents the semantics of the Bulgarian verb); transfer (applies when the lexical unit is included only in the Bulgarian FrameNet, and the semantic frame is assigned based on an inheritance relation); modification (when an adaptation of a semantic frame is required to comply with the Bulgarian data) and creation of new semantic frames; and d) linking the core frame elements with nouns' semantic classes or a combination of classes ensuring the expression of verb-noun syntagmatic relations. BulFrame, a web-based system designed for creating, editing, validating and viewing semantic frames, is also presented. Data from the Bulgarian FrameNet will be used in future studies for semantic role labelling tasks.

Dr. Oliver Czulo is full professor of Translation Studies at the University of Leipzig. Until March 2017, he held an Assistant Professor position for Translation-relevant Linguistics at the Translation Faculty at the University of Mainz. He's a member of the European Society for Translation Studies and a founding member of the Global FrameNet initiative.

Seminar: Do we need an open access book series for Global FrameNet?
Date: November 1st, 2022
Abstract: Framenet projects all over the world have been very active not only in the creation of framenets, but also in neighbouring fields such as NLP where they have contributed to research on annotating and modelling semantics. Recently, cross-fertilization with multimodal research or translation studies has also gained momentum. These activities and their outcomes meet a publishing landscape that has seen a strong move towards open access publishing. In this sessions, I will propose to found an open access book series devoted to the dissemination of "Global perspectives on frames (and constructions?)" (working title) at the open access publisher Language Science Press. I will inform on the most important aspects of publishing with LSP so that after the sessions the picture of whether this is an endeavour worth pursuing should be clearer.

Debanjana Kar is a Research Engineer at IBM Research and her research interests broadly include NLP and Deep Learning. Prior to IBM Research, she completed her Masters from IIT Kharagpur specialising in Multilingual Information Extraction and has experience working in varied NLP domains such as text generation, text classification, machine translation. She has published her research works at top-tier conferences like PAKDD, AACL-IJCNLP, ACL. She is also associated with the Global FrameNet organisation and has been a mentor at GSoC for the last few years after a successful stint as a student developer at the same.

Seminar: Exploring MT Evaluation using Frame Semantic Parsing.
Date: December 6th, 2022
Abstract: Evaluation techniques adopted for the tasks of Natural Language Generation and Machine Translation have been strongly scrutinized and debated for a very long time in the NLP community. Popularly used evaluation metrics have often been deemed unable to capture adequacy and fluency in the machine outputs resulting in poorer correlations with human references. While more recent techniques help capture adequacy and fluency better, we find the lack of a metric which explicitly evaluates both the syntactic and semantic qualities of the machine output, with or without a human-annotated reference. In this talk, we will discuss recent works in the domain of machine translation evaluation along with a discussion of our own frame-based translation evaluation system that builds on the gaps of the existing works. We will present a proof-of-concept of a frame-based  metric that assesses both adequacy & fluency of a machine translation or generation output.