Varied Applications of Ontology for Vertical Search, Information Visualization, and Data Fusion
Abstract:
We are testing semantic means to improve information access and display. Ontologies help perform standard classifications in a prototype information system for paleontology, but more interestingly, they underlie our geovisualization, and we are experimenting with ontologies to sub-classify results retrieved in nested-rectangle, tree map form. Ontologies in a prototype biomedical information system help find data overlaps for subsequent manual verification. Both systems are still under development, but results look promising.
The paleontology system is joint work with Dr. Carbonell, Dong Cao, Raymond Lu and Laura Meng at Carnegie Mellon, paleontologist Dr. Chris Noto, and the GEON project in cloud computing for the geosciences. The biomedical project is in concert with Dr. Lesk at Rutgers University and Drs. Tarter, Vanyukov, Kirisci and Ridenour of the Center for Education and Drug Abuse Research (CEDAR) at the University of Pittsburgh.
Bio:
Judith Gelernter is a postdoc in the Language Technologies Institute. She received a PhD in information science in 2008 from Rutgers University. A revised version of her dissertation has just been published by VDM Verlag as "Intelligent Information Retrieval for Maps." She looks for ways to improve information system access via back end semantic means and front end interface design.
May 15 Dan Klein
UC Berkeley
Joint LTI/Intelligence Seminar
Forest-Based Algorithms in Natural Language Processing
Abstract:
Language is complex, but our labeled data sets generally aren't. For example, treebanks specify coarse categories like noun phrases, but they say nothing about richer phenomena like agreement, case, definiteness, and so on. One solution is to use latent-variable methods to learn these underlying complexities automatically. In this talk, I will present several latent-variable models for natural language processing which take such an approach.
In the domain of syntactic parsing, I will describe a state-splitting approach which begins with an X-bar grammar and learns to iteratively refine grammar symbols. For example, noun phrases are split into subjects and objects, singular and plural, and so on. This splitting process in turn admits an efficient coarse-to-fine inference scheme, which reduces parsing times by orders of magnitude. Our method currently produces the best parsing accuracies in a variety of languages, in a fully language-general fashion. The same techniques can also be applied to acoustic modeling, where they induce latent phonological patterns.
In the domain of machine translation, we must often analyze sentences and their translations at the same time. In principle, analyzing two languages should be easier than analyzing one: it is well known that two predictors can work better when they must agree. However "agreement'' across languages is itself a complex, parameterized relation. I show that, for both parsing and entity recognition, bilingual models can be built from monolingual ones using latent-variable methods -- here, the latent variables are bilingual correspondences. The resulting bilingual models are substantially better than their decoupled monolingual versions, giving both error
rate reductions in labeling tasks and BLEU score increases in machine translation.
Bio:
Dan Klein is an assistant professor of computer science at the University of California, Berkeley (PhD Stanford, MSt Oxford, BA Cornell). His research focuses on statistical natural language processing, including unsupervised learning methods, syntactic parsing, information extraction, and machine translation. Academic honors include a Marshall Fellowship, a Microsoft New Faculty Fellowship, the ACM Grace Murray Hopper award, and best paper awards at the ACL, NAACL, and EMNLP conferences.
May 14 Liang Huang
Google Research
Forest-Based Algorithms in Natural Language Processing
Abstract:
Many problems in Natural Language Processing (NLP) involves an efficient search for the best derivation over (exponentially) many candidates. For example, a parser aims to find the best syntactic tree for a given sentence among all derivations under a grammar, and a machine translation (MT) decoder explores the space of all possible translations of the source-language sentence. In these cases, the concept of packed forest provides a compact representation of huge search spaces by sharing common sub-derivations, where efficient algorithms based on Dynamic Programming (DP) are possible.
This talk first develops fast and exact k-best DP algorithms on forests, which are orders of magnitudes faster than previously used methods on state-of-the-art parsers. We also show empirically how the improved output of our algorithms has the potential to improve results from parse reranking systems and other applications.
We then extend these algorithms for approximate search when the forests are too big for exact inference. We discuss two particular instances of this new method, forest rescoring for MT decoding, and forest reranking for parsing. In both cases, our methods perform orders of magnitudes faster than conventional approaches. In the latter, faster search also leads to better learning, where our approximate decoding makes whole-Treebank discriminative training practical and results in an accuracy better than any previously reported systems trained on the Treebank.
Finally, we also apply the forest concept to syntax-based translation, where we use the packed forest, instead of 1-best tree, to direct a tree-based decoding process, and to extract translation rules. Large-scale experiments show ~2.5 BLEU points of improvement over the 1-best baseline, which enables our forest-based system to significantly outperform the state-of-the-art hierarchical system Hiero.
(This is an overview of my Ph.D. work, and includes joint work with David Chiang, Kevin Knight, Aravind Joshi, Haitao Mi, and Qun Liu.)
Bio:
Liang Huang is a Research Scientist at Google Research, Mountain View, CA. He received his PhD from the University of Pennsylvania in 2008, co-supervised by Aravind Joshi and Kevin Knight (USC/ISI). He is mainly interested in the theoretical aspects of computational linguistics, in particular, efficient algorithms in parsing and machine translation, generic dynamic programming, and formal properties of synchronous grammars. His work received an Outstanding Paper Award at ACL 2008, and Best Paper Nominations at ACL 2007 and EMNLP 2008.
April 10
Qin Jin
LTI
Speaker De-Identification via Voice Transformation
Abstract:
Speaker identification might be a suitable answer to prevent unauthorized access to personal data. However we also need to provide solutions to secure transmission of spoken information. This challenge divides into two major aspects. First, the secure transmission of the content of the spoken input and second the secure transmission of the identity of the speaker. In this talk, we present our investigation concentrating on the latter, i.e. how to securely transmit information via voice without revealing the identity of the speaker to un-authorized listeners. In order to make the first steps toward solving this problem we study the potential of voice transformation for speaker de-identification. We use two speaker identification approaches to verify the success of de-identification with voice transformation, a GMM-based and a Phonetic approach, and study different voice transformation strategies to disguise speaker identity information while preserving understandability.
Biography:
Qin Jin received her PhD from Language Technologies Institute (LTI), Carnegie Mellon University (CMU) in 2007, with a thesis title "Robust Speaker Recognition". She is now a faculty member at LTI CMU. Qin Jin¡¯s research interest includes robust speaker recognition, human biometrics, and speech recognition and understanding.
March 20
Florian Metze
LTI
On using Articulatory Features for Discriminative Speaker Adaptation
Abstract:
This talk presents a way to perform speaker adaptation for automatic speech recognition using the stream weights in a multi-stream setup, which included acoustic models for "Articulatory Features" such as "Rounded" or "Voiced". We present supervised speaker adaptation experiments on a spontaneous speech task and compare the above stream-based approach with experiments on model-based adaptation. In the approach we present, stream weights model the importance of features, which offers a descriptive interpretation of the adaptation parameters.
Biography:
Florian Metze received his PhD from Universität Karlsruhe (TH) in 2005, with a thesis on "Articulatory Features for Conversational Speech Recognition". After working between academia and industry as a post-doc with Deutsche Telekom Laboratories in Berlin, he is now faculty at Carnegie Mellon's LTI. His research interests are in the areas of acoustic modeling, particularly for distant microphone speech, speech processing applications, and semantic computing.
March 5
David Traum
University of Southern California
Multi-party, Multi-issue, Multi-strategy Negotiation for Multi-modal Virtual Agents
Abstract:
We present a dialogue model of face to face negotiation for virtual agents that extends previous work to be more human-like and applicable to a broader range of situations, including more than two negotiators with different goals, and negotiating over multiple options. The agents can dynamically change their negotiating strategies based on the current values of several parameters and factors that can be updated in the course of the negotiation. We have implemented this model and done preliminary evaluation within a prototype training system and a three-party negotiation with two virtual humans and one human.
Biography:
Dr. David Traum (http://www.ict.usc.edu/~traum) is a Research Scientist at ICT and a research assistant professor of Computer Science at the University of Southern California. He completed his Ph.D. in Computer Science at the University of Rochester in 1994. His research focuses on collaboration and dialogue communication between agents, including both human and artificial agents. Of primary interest is the interaction between the individual cognitive functioning and the social fabric, and the relationship between task-related and communicative actions. He has engaged in theoretical, implementational, and empirical approaches to the problem, studying human-human natural language and multi-modal dialogue, as well as building a number of dialogue systems to communicate with human users. These systems have ranged in complexity from simple command and control and system-directed information-providing systems to full mixed-initiative collaborative planning and interaction, and have included both uni-modal (text or speech) and multimodal (gesture, sketching, pointing) systems and embodied conversational agents. One major thrust of research includes the ?rounding?problem ?how do communicators realize how well they are understanding each other, and what steps can and should they take to increase this mutual understanding (including giving positive and negative feedback, in a variety of modalities)? Dr. Traum is author of over 150 technical articles, has served on many conference program committees, and is currently the president emeritus of SIGDIAL, the international special interest group in discourse and dialogue.
February 20
Agustin Gravano
Columbia University
A model of turn-taking in task-oriented dialogue
Abstract:
As interactive voice response systems spread at a rapid pace, providing
an increasingly more complex functionality, it is becoming clear that
the challenges of such systems are not solely associated to their
synthesis and recognition capabilities. Rather, issues such as the
coordination of turn exchanges between system and user appear to play an
important role in system usability. This study explores that issue
in the Columbia Games Corpus, a collection of spontaneous task-oriented
dialogues in Standard American English. We provide evidence of the
existence of seven turn-yielding cues -- prosodic, acoustic and
syntactic events strongly associated with conversational turn endings --
and show that the likelihood of a turn-taking attempt from the
interlocutor increases linearly with the number of cues conjointly
displayed by the speaker. We present similar results related to six
backchannel-inviting cues -- events that invite the interlocutor to
produce a short utterance conveying continued attention.
Biography:
Agustin Gravano recently defended his Ph.D. thesis in the Computer
Science Department at Columbia University, New York. In 2001, he had
earned his B.S. in Computer Science from the University of Buenos Aires,
Argentina, where he will be returning in July 2009. His main research
topic is prosodic variation in spoken dialogue, aimed at improving the
models used in interactive voice response systems, both for understading
the user's input and for generating natural responses.
December 12
Ni Lao
Language Technologies Institute, CMU
Robust Cross-Lingual Information Retrieval
Abstract:
In this talk, I will describe the Information Retrieval subsystem of JAVELIN IV, a question-answering system that answers complex questions from multilingual sources. I will focus on different strategies for query term extraction, translation, filtering, expansion and weighting, including a alias expansion technique using lexico-syntactic patterns learned with weakly-supervised algorithm. In the NTCIR7 IR4QA evaluation, our retrieval system achieved 59% and 59% MAP in the Chinese-to-Chinese and Japanese-to-Japanese subtasks, respectively. We provide a rationale for the retrieval system design, and present a detailed analysis for the effect of machine translation on IR performance.
December 12
Hideki Shima
Language Technologies Institute, CMU
Complex Cross-lingual Question Answering as a Sequential
Classification and Multi-Document Summarization Task
Abstract:
Current research in Question Answering is shifting toward answering
"complex" non-factoid questions. Many systems end up in sentence
retrieval to narrow down the search space. However, there is a
limitation in this approach, since not all answer-bearing passage
contains the query term in the question. We discuss our novel answer
extraction approach, which views the problem as a sequential
classification task, together with the use of different units of
extraction, and the effect of different features for classification.
We then describe the answer generation algorithm inspired by
multi-document summarization task. Our talk is based on the CMU
JAVELIN system evaluated in the NTCIR-7 ACLIA Complex Cross-lingual QA
task with four complex question types: definitions, biographies,
relationships, events and four language pairs: English-to-Japanese,
Japanese-to-Japanese, English-to-Chinese, Chinese-to-Chinese.
The system achieved the best score among participants in
English-to-Japanese and Japanese-to-Japanese subtasks.
November 21
Ray Mooney
University of Texas at Austin
Learning Language from its Perceptual Context
Abstract:
Current systems that learn to process natural language require
laboriously constructed human-annotated training data. Ideally, a
computer would be able to acquire language like a child by being
exposed to linguistic input in the context of a relevant but ambiguous
perceptual environment. As a step in this direction, we present a
system that learns to sportscast simulated robot soccer games by
example. The training data consists of textual human commentaries on
Robocup simulation games. A set of possible alternative meanings for
each comment is automatically constructed from game event traces. Our
previously developed systems for learning to parse and generate
natural language (KRISP and WASP) were augmented to learn from this
data and then commentate novel games. The system is evaluated based
on its ability to parse sentences into correct meanings and generate
accurate descriptions of game events. Human evaluation was also
conducted on the overall quality of the generated sportscasts and
compared to human-generated commentaries.
Biography:
Raymond J. Mooney is a Professor in the Department of Computer Sciences at the
University of Texas at Austin. He received his Ph.D. in 1988 from the
University of Illinois at Urbana/Champaign. He is an author of over 150
published research papers, primarily in the areas of machine learning and
natural language processing. He is the current President of the International
Machine Learning Society, was program co-chair for the 2006 AAAI Conference on
Artificial Intelligence, general chair of the 2005 Human Language Technology
Conference and Conference on Empirical Methods in Natural Language Processing,
and co-chair of the 1990 International Conference on Machine Learning. He is a
Fellow of the American Association for Artificial Intelligence and recipient
of best paper awards from the National Conference on Artificial Intelligence,
the SIGKDD International Conference on Knowledge Discovery and Data Mining,
the International Conference on Machine Learning, and the Annual Meeting of
the Association for Computational Linguistics. His recent research has focused
on learning for natural-language processing, text mining for bioinformatics,
statistical relational learning, and transfer learning.
November 7
Gregory Aist
Arizona State University
Computing with Language and Context over Time
Abstract:
How do language and context interact in learning and performance by humans and machines? To explore this broad area of inquiry, I have studied interactions between natural language and a wide range of different contexts: visual context, social and team context, written context and world knowledge, procedure and task context, dialogue and temporal context, and instructional context. Specific research questions have included how machines can process spoken language continuously and integrate speech and visual context during understanding; how computers can help pilots and astronauts learn and perform tasks; and how to automatically generate, present, and evaluate the effects of vocabulary help for children. One key challenge in addressing all of these questions is to model and compute representations of language and context that unfold over time as the interaction progresses. This talk will illustrate the need for such interactive time-sensitive processes,
describe computational approaches to understanding language and context as dialogue and interactions unfold across time, and evaluate the effectiveness of such approaches.
Biography:
Gregory Aist is currently at Arizona State University as an Assistant Research Professor in the School of Computing and Informatics and the Applied Linguistics Program. His research interests are in natural language processing and computer-assisted learning. His research addresses fundamental issues in language and learning, tackles computational challenges of automatic processing of human language and computer support for human learning, and is applied to provide users with learning experiences and new capabilities in authentic settings for educational domains such as traditional literacy (reading and writing) and new literacies (virtual worlds), and physical domains such as aerospace and human-robot interaction. During summers 2007 and 2008 he was an Air Force Summer Faculty Fellow. Previously he has held research and visiting positions at the University of Rochester, RIACS/NASA Ames Research Center, and the MIT Media Lab. He received a Ph.D. in Language and Information Te!
chnology from Carnegie Mellon University in 2000, where he was an NSF Graduate Fellow.
October 31
Pedro Domingos
University of Washington
From Text to Knowledge via Markov Logic
Abstract:
Language understanding is hard because it requires a lot of knowledge.
However, the only cost-effective way to acquire a lot of knowledge is
by extracting it from text. The best (only?) hope for solving this
"chicken and egg" problem is bootstrapping: start with a small knowledge
base, use it to process some text, add the extracted knowledge to the KB,
process more text, etc. Doing this requires a modeling language that can
incorporate noisy knowledge and seamlessly combine it with statistical
NLP algorithms. Markov logic accomplishes this by attaching weights to
first-order formulas and viewing them as templates for features of Markov
random fields. In this talk, I will describe some of the main inference
and learning algorithms for Markov logic, and the progress we have made
so far in applying them to NLP. For example, we have developed a system
for unsupervised coreference resolution that outperforms state-of-the-art
supervised ones on MUC and ACE benchmarks.
Biography:
Pedro Domingos is Associate Professor of Computer Science and
Engineering at the University of Washington. His research interests
are in artificial intelligence, machine learning and data mining. He
received a PhD in Information and Computer Science from the University
of California at Irvine, and is the author or co-author of over 150
technical publications. He is a member of the advisory board of JAIR,
a member of the editorial board of the Machine Learning journal, and a
co-founder of the International Machine Learning Society. He was
program co-chair of KDD-2003, and has served on numerous program
committees. He has received several awards, including a Sloan
Fellowship, an NSF CAREER Award, a Fulbright Scholarship, an IBM
Faculty Award, and best paper awards at KDD-98, KDD-99 and PKDD-2005.
October 17
Larry Heck
Search & Advertising Sciences, Yahoo! Labs
An Introduction to Yahoo! Labs
Abstract:
Yahoo! Labs is the central advanced research and development organization of Yahoo! Inc., a leading global Internet brand and one of the most trafficked Internet destinations worldwide. We're responsible for innovating the Internet: all the way from the big invention through to deploying that invention into production. Our goals are nothing short of inventing the future of the Internet and creating the next generation of businesses for Yahoo! In this talk, I will give an overview of Yahoo! Labs and provide a glimpse into the technical problems we are working hard to solve.
Biography:
Dr. Larry Heck is Vice President of Search & Advertising Sciences at Yahoo. In this role, Dr. Heck leads the teams responsible for the scientific development and deployment of Yahoo search and monetization algorithms. He received the PhD in Electrical Engineering from the Georgia Institute of Technology in 1991. He then worked at SRI International and served as principal investigator for a number of federally funded research programs (NSA, DARPA, ORD/CIA) in acoustics and speech, including active noise & vibration control, acoustic machinery monitoring, and speaker recognition. After SRI, Dr. Heck was Vice President of R&D at Nuance Communications, where he led teams responsible for natural language processing, speech recognition, voice authentication, and text-to-speech synthesis. Dr. Heck has published over 50 scientific articles, and has served on numerous boards for the IEEE and International Speech Communication Association.
October 16
Ee-Peng Lim
Singapore Management University
On Evaluating and Mining Online Rating Data
Online rating system is a popular feature of Web 2.0 applications. It typically involves a set of reviewers assigning rating scores (based on various evaluation criteria) to a set of objects. We identify two objectives for research on online rating data, namely achieving effective evaluation of objects and learning behaviors of reviewers/objects. These two objectives have conventionally been pursued separately. In our research, we have attempted to address both objectives concurrently. We introduce models for several interesting behaviors, i.e., user bias and leniency and object's controversy and quality. We have developed these models using the dependency relationships among them. Finally, we evaluate and compare the behavior models through experiments on real and synthetic data.
Biography: Ee-Peng Lim is currently a tenured professor at the School of Information Systems of the Singapore Management University (SMU). He received Ph.D. from the University of Minnesota, Minneapolis in 1994 and B.Sc. in Computer Science from National University of Singapore. His research interests include information integration, data/text/web mining, and digital libraries.He is currently an Associate Editor of the ACM Transactions on Information Systems (TOIS), Journal of Web Engineering (JWE), International Journal of Digital Libraries (IJDL) and International Journal of Data Warehousing and Mining (IJDWM). He is a member of the ACM Publications Board.He is the Steering Committee Chair of the International Conference on Asian Digital Libraries (ICADL), and a member of the Steering Committee of the Pacific Asia Conference on Knowledge Discovery and Data Mining (PAKDD).
Sept 25
Roland Hausser
Universitaet Erlangen-Nuremberg, Germany
Linguistic Methods for Improving Automatic Document Search
Abstract:
Today's search engines build their indices on the basis of document
mark-up in XML and significant letter sequences (words) occurring in
the document texts. There are some drawbacks, however: the XML
mark-up requires skill as well as tedious work from the user posting
the document, and the indexing based on significant word
distributions, though automatic and highly effective, is not as
precise as required by many applications.
As a complement to current methods, this paper presents an automatic
content analysis of texts which is based on traditional linguistic
methods in conjunction with the comparatively new data type of
proplets and the algorithm of LA-grammar. The idea is to use the
grammatical relations of functor-argument structure and coordination
to refine the indexing for storage and retrieval. The approach will
be illustrated by an implementation in Java.
Biography:
Professor Roland Hausser heads the Department of
Computational Linguistics at the University of Erlangen-Nuremberg in
Germany. He has worked in formal semantics, grammar algorithms, and
human-computer communication. Publications include several books on
Database Semantics, the latest `A Computational Model of Natural
Language Communication,' as well as papers in the AI Journal and the
Journal of Theoretical Computer Science.
