FrameNet Workshop Speakers and Sessions

←  Back to main event page

Miriam R. L. Petruck

Introduction to FrameNet Semantics

Monday, September 9, 9:00 - 9:30

Intro to Frame Semantics and the FrameNet Knowledge Base

Tuesday, September 10, 9:00 - 10:30

FrameNet Annotation

Tuesday, September 10, 15:45 - 17:00

Bio:

Having written the first dissertation in Frame Semantics and the lexicon under the supervision of Charles J. Fillmore, Miriam R. L. Petruck has been contributing to the development of the theory from before Fillmore's founding of FrameNet. She has published numerous papers demonstrating the efficacy of Frame Semantics for the characterization of the lexicon, and about FrameNet. Petruck received the Ph.D. in Linguistics from the University of California, Berkeley, and currently is a member of the Aritificial Intelligence group at ICSI working primarily on FrameNet.

Collin Baker

The FrameNet Knowledge Base

Monday, September 9, 9:30 - 10:00

FrameNet Representations and Their Significance for Inferencing

Monday, September 9, 10:00 - 10:30

Applications and Extensions of FrameNet

Monday, September 9, 11:15 - 11:45

Approaches to Expanding FrameNet

Monday, September 9, 14:45 - 15:15
Friday, September 13, 14:30 - 15:30

Intro to Frame Semantics and the FrameNet Knowledge Base

Tuesday, September 10, 9:00 - 10:30

FrameNet Data and the Representation of Events in FrameNet

Tuesday, September 10, 10:45 - 12:15

Bio:

Collin F. Baker has been Project Manager of the FrameNet project at ICSI since 2000, working in close collaboration with its founder, Charles J. Fillmore, Prof. Emeritus of Linguistics at UC Berkeley, supervising day-to-day operation and helping with data structures and software design.  He has been a member of the FrameNet team since its inception in 1997, and received his Ph.D. in Linguistics from UC Berkeley in 1999; he is also project manager of the MetaNet project at ICSI.  He has published numerous papers about FrameNet as a semantically rich knowledge base from both linguistic and computational perspectives, and has served as a reviewer for NSF (CISE) and for journals such as Computational Linguistics, Language Resources and Evaluation, Cognitive Linguistics, Constructions and Frames, and Cognitive Science. In 2012, jointly with Prof. Fillmore, he was a recipient of the Antonio Zampolli Prize from the European Language Resources Association; in 2013, he received a Google Research Award for a project entitled "Scaling Up FrameNet for NLP".

Nancy Chang

From Frames to Inference: Event Representation in FrameNet and Beyond

Monday, September 9, 10:45 - 11:15
Tuesday, September 10, 10:45 - 12:15 (with Collin Baker)

This talk explores how rich event representations motivated by sensorimotor constraints can be used for language understanding and inference. I present a framework in which language understanding draws on embodied simulations of actions and events, as modeled by a dynamic computational formalism. Linguistic structures, on this view, serve to specify which structures are involved in the simulation and how they are related; the simulation then yields rich, context-sensitive inferences that constitute the meaning of an utterance. I describe the basics of simulation semantics and present several applications illustrating the power and flexibility of this approach for addressing key challenges in natural language understanding, including aspectual inference, perspective and question answering.

Bio:

Nancy Chang is a software engineer at Google working on conversational search and natural language understanding at Google. After earning her doctorate in Computer Science at UC Berkeley, she served as a research associate at the Sony Computer Science Laboratory and the Université Sorbonne Nouvelle-Paris 3 and as a visiting lecturer at Gothenburg University. Her research focuses on computational models of language structure, learning and use.

Dipanjan Das, Google

Automatic Semantic Role Labeling for FrameNet Frame Elements

Monday, September 9, 13:15 - 13:45
Tuesday, September 10, 13:15 - 14:30

In this talk, I will focus on data-driven models for semantic structure prediction following frame semantics (Fillmore, 1982), a linguistic theory that describes predicate-argument relationships and emphasizes the abstraction of predicate meaning into semantic frames. Our method exploits rich information provided by linguists in the form of a lexicon (Fillmore and Baker, 2010, Ruppenhofer et al., 2010), as well as a small amount of annotated data, to automatically find disambiguated semantic frames of lexical predicates present in a sentence. A frame represents semantic knowledge and requires semantic roles that are fulfilled by arguments, in the form of words and phrases within the sentence. After disambiguating each predicate to the frame it evokes, our method finds the frame's arguments collectively via joint inference, making use of dual decomposition. Large amounts of annotated data for this task are unavailable; to this end, we model latent structure and apply semi-supervised learning, resulting in more robust models with broader coverage. Frame semantics is richer than the representation used in popular semantic role labeling systems (Kingsbury and Palmer, 2002) but less domain-specific than semantic parsers based on logical form (Ge and Mooney, 2005; Zettlemoyer and Collins, 2005); it represents a viable "middle ground" for data-driven semantic analysis of text. Compared to previous work, our method makes fewer independence assumptions and significantly outperforms past state of the art.

Bio:

Dipanjan Das is a research scientist at Google in New York. He works on statistical natural language processing, focusing on semi-supervised learning of syntax and semantics. He completed his M.S. and Ph.D. from the Language Technologies Institute, School of Computer Science at Carnegie Mellon University in 2008 and 2012 respectively. Das completed his undergraduate degree in Computer Science and Engineering in 2005 from the Indian Institute of Technology, Kharagpur. His research in cross-lingual projection of syntax received the best paper award at the ACL 2011 conference.

Tim Hawes, Decisive Analytics Corporations

Real World Text Analytics Using FrameNet

Monday, September 9, 13:45 - 14:15
Wednesday, September 11, 9:00 - 10:30

When it comes to text, big data often means a massive diversity of ideas. Providing tools to users that will let them work with these ideas instead of keywords, which can be ambiguous and incomplete identifiers of specific concepts, is a challenging problem. In this talk, I will discuss some of the ways we are addressing this diversity of ideas with FrameNet based technologies at Decisive Analytics Corporation, including: Attitude Analysis, which can provide a perspective on how groups discussed in a corpus relate to one another; Network Extraction and Refinement techniques that identify relationships between entities and convert the resulting “hairball” into an interpretable graph; and a Semantic Search tool which allows users to query data based on concepts rather than keywords.

Bio:

Tim Hawes is a scientist for Decisive Analytics Corporation’s Analytic Products Division. He has over nine years of experience in Natural Language Processing with a background in computational, theoretical and psycho- linguistics, and computer science. Mr. Hawes holds a B.S. (Northeastern University) and M.A. (University of Maryland, College Park) in Linguistics. His graduate research concentrated on building predictive models of U.S. Supreme Court voting patterns from linguistic behavior. Mr. Hawes has led the development of DAC’s advanced sentiment analysis capability.

Josef Ruppenhofer, University of Hildesheim

Sentiment Analysis Using FrameNet

Monday, September 9, 14:15 - 14:45
Wednesday, September 11, 10:45 - 12:15

In this presentation, I will show what FrameNet offers sentiment analysis. In particular, I focus on the case of fine-grained analysis of individual sentiment-bearing expressions. This is relevant for, e.g., opinion Q&A or summarization where we do not merely want to know, is this  document (mostly) objective or neutral, or is it positive or negative in tone, but we want to know specifically who [source] feels how strongly [intensity] positive or negative [valence/(polarity] about what [target]?

First and foremost, FrameNet allows piggy-backing a good deal of the source and target identification problem onto the task of (automatic) semantic role labeling. FrameNet covers many emotion- and communication-related frames that are necessary to do this work. Furthermore, FrameNet contains not senses of multi-word expressions, not only single words, and even accommodates syntactic patterns and constructions through the FrameNet constructicon. In addition, the FrameNet knowledge base already has positive/negative markings  on particular (antonymous) lexical units that indicate polarity/valence.

However, some types of information are missing from the FrameNet knowledge base, but could be added. For instance, we would like to add information on scalar intensity, indicating, for instance that handsome < gorgeous. I will show possible extensions to FrameNet's representation to accommodate these new types of information and discuss ways to acquire this information semi-automatically.

Jerome Feldman

The MetaNet Project at ICSI

Monday, September 9, 15:30 - 16:15

FrameNet has long held the goal of including information about metaphorical usage in language. The MetaNet project, based at ICSI, is a multi-lingual, multi-site, multi-disciplinary metaphor processing undertaking that incorporates FN methodology as well as corpus and machine learning techniques, deep cognitive linguistics, and behavioral and imaging experiments. This talk will describe the overall structure of the project and the resulting repository structure, which has the potential to support a wide range of applications.

Bio:

Over a span of five decades, Jerome Feldman has made significant contributions to many areas of computer science and built important systems in cutting edge areas. His early work on compilers and associative programming languages was influential and is generally cited as one of the foundations of relational data bases. At Stanford, he was associate director of the AI lab and led the vision and robotics effort, which has had a profound impact on all aspects of automation. Upon founding the Rochester CS Department, he led a project on distributed operating systems that established the basis for much of the technology in current use. He was also one of the initiators of the connectionist approach to AI and Cognitive Science.

Upon coming to UC Berkeley as founding director of ICSI, he established connectionist (neural) computation as a cornerstone of ICSI and this remains central. For the last two decades, he has been collaborating with Prof. George Lakoff and others on the Neural Theory of Language (NTL), which is the topic of his recent MIT Press book From  Molecule to Metaphor. He has supervised over a hundred doctoral and post-doctoral students who have gone on to leading positions in academia, industry, and government.

Eve Sweetser

Cognitive Linguistic Aspects of Metaphor

Friday, September 13, 9:00 - 9:30

Metaphor involves mapping between two different cognitive/semantic FRAMES or SCHEMAS. The MetaNet Wiki is a database of such mappings, drawing on FrameNet's inventory of frames. Although the focus of the project is metaphoric concepts of economics and government, many other basic metaphoric structures (e.g., for action and causation) have necessarily been mapped as parts of that more specific database. Systematic cognitive structures underlie the metaphoric linguistic usages found in texts about economics and politics - no individual word is a "metaphor," but a particular word usage can be a manifestation of such a cognitive metaphoric mapping - which may be systematically manifested in many different linguistic uses.

Bio:

Eve E. Sweetser is Professor of Linguistics at the University of California, Berkeley. Her primary research interests include semantics and meaning changes, the semantics of grammatical constructions, cognitive linguistics, metaphor and iconicity, subjectivity and viewpoint, the relationship between language and gesture, historical linguistics, and the Celtic language family. Her 1990 book, From Etymology to Pragmatics (Cambridge University Press), explores generalizations about synchronic and diachronic patterns of meaning in the areas of model verbs and conjunctions. Her 2005 book, Mental Spaces in Grammar: Conditional Constructions, was coauthored with Barbara Dancygier and examines the syntax and semantics of a wide range of English conditional constructions, using a Mental Spaces model of semantics.

Katia Shutova

Computational Modeling of Metaphor Using Unsupervised Learning

Friday, September 13, 9:30 - 10:00

Metaphor processing is a rapidly growing area in NLP. The pervasiveness of metaphor in language and the role that metaphor plays in human reasoning makes its automatic identification and interpretation indispensable for many practical NLP applications. The recent rise of statistical approaches in the field of metaphor processing opened new avenues for increasing system accuracy and robustness. In this talk, I will present our recent work applying unsupervised learning techniques to the problem of metaphor identification. Our system discovers both metaphorical associations and metaphorical expressions in unrestricted text. It first performs hierarchical graph factorization clustering (HGFC) of nouns and then searches the resulting graph for metaphorical connections between concepts (e.g., ARGUMENT is a WAR). It then makes use of the salient features of the metaphorically connected clusters to identify the actual metaphorical expressions (e.g., He shot down all of my arguments). Aside from performing with a promising recall and precision, this approach is the first in NLP to exploit the cognitive findings about differences in the organization of abstract and concrete concepts in the human brain.

Bio:

Katia Shutova is a Postdoctoral Research Fellow at the International Computer Science Institute and the Institute for Cognitive and Brain Sciences, University of California, Berkeley. Her research focuses on computational modelling of human creativity and metaphor. She currently leads the  Metaphor Extraction research team at ICSI, whose goal is to create robust and accurate tools that identify metaphorical expressions in unrestricted text using statistical methods. Before coming to Berkeley, she was a Research Associate at RCEAL and the Computer Laboratory, University of Cambridge, working on issues in computational lexical semantics. She received her PhD from the University of Cambridge Computer Laboratory in 2011. Katia also holds an MPhil from the University of Cambridge and an undergraduate degree in computational linguistics from St-Petersburg State University.

Oana David, Ellen Dodge, Jisup Hong, and Elise Stickles

MetaNet Demo

Friday, September 13, 10:15 - 12:15

The MetaNet Repository is a functional repository and knowledge base of conceptual metaphors, schemas, and linguistic examples of conceptual metaphors.  The repository is designed primarily as a computational resource for automatic metaphor detection and interpretation.  However, as advances in conceptual metaphor theory have significantly increased the complexity of metaphor analyses, it is also designed to be a platform for multi-lingual metaphor research.   We present various parts of this system in the form of a demo, including the metaphor ontology, the structure of the repository, the semantic wiki used by linguistic and conceptual analysts to build conceptual networks, and how the system is used to connect automatically extracted examples to conceptual metaphors.

Bio:

Jisup Hong is Postdoctoral Research Fellow at the International Computer Science Institute, where he works jointly with the analysis and repository teams in the MetaNet project to develop formal representations for conceptual metaphor theory, and to build metaphor repositories as resources for automatic interpretation.  He completed his PhD in Linguistics at U.C. Berkeley in 2012, and holds a BA in Computer Science from Harvard University.

Gerald Friedland

FrameNet and Privacy

Friday, September 13, 13:15 - 14:30

We have shown [1] that potential attackers can identify accounts on different social network sites that all belong to the same user, exploiting only innocuous activity that inherently comes with posted content, such as writing style.  In a study we examined three specific features on Yelp, Flickr, and Twitter: the geo-location attached to a user’s posts, the timestamp of posts, and the user’s writing style as captured by language models. By combining all three features, the accuracy of identifying Twitter accounts that belong to a set of Flickr users is comparable to that of existing attacks that exploit usernames. Furthermore, our attack can identify 37% more accounts than using usernames when we instead correlate Yelp and Twitter. These results have significant privacy implications as they present a novel class of attacks that exploit users’ tendency to assume that, if they maintain different personas with different names, the accounts cannot be linked together; whereas we show that the posts themselves can provide enough information to correlate the accounts. In order to mitigate attacks like this, one strategy is to blur location and time-stamps. Blurring writing style, however, requires an understanding of the semantics of the social network posts. This talk will present the study and explore ideas on blurring writing styles using FrameNet and other tools.

[1] Oana Goga, Howard Lei, Sree Hari Krishnan Parthasarathi, Gerald Friedland, Robin Sommer, and Renata Teixeira. 2013. Exploiting innocuous activity for correlating users across sites. In Proceedings of the 22nd international conference on World Wide Web (WWW '13). International World Wide Web Conferences Steering Committee, Rio de Janeiro, Brazil, 447-458.

Bio:

Gerald Friedland studied at Freie Universitaet in Berlin, receiving his master’s degree in 2002 and his doctorate summa cum laude in 2006. He moved to Berkeley for a two-year postdoctoral fellowship at ICSI funded by the German Academic Exchange Service (DAAD). The fellowship was extended with additional funds from Appscio Inc. In 2008, he accepted a position as a senior researcher at ICSI. He has received the European Academic Software Award and the IEEE Computer Society Distinguished Service Award. In 2011, he was named associate editor of the year by ACM Transactions on Multimedia Computing, Communications, and Applications. He is a co-founder of the IEEE International Conference on Semantic Computing and the IEEE Summer School of Semantic Computing and a program co-chair of the International Conference on Multimedia and Expo 2012. In 2009 he won the ACM Multimedia Grand Challenge and in 2011 served as its chair. He leads ICSI’s research efforts in multimedia understanding; his work focuses on audio techniques such as speaker diarization and acoustic event detection and their applications to multimedia content analysis.