Download presentations / posters / papers

Keynotes (abstracts):

Dr. Fadi Biadsy from Google Research New York will talk about speech to text with emphasis on Hebrew and Arabic.
Dr. Guy Hoffman from IDC will talk about human-robot interaction.

Dr. Yoelle Maarek from Yahoo! Labs will talk about "Queries and Questions in Yahoo! Answers"

See the posters

Tentative Schedule:


09:00 - 09:30
Coffee and registration

09:30 - 09:40
Opening words and annoucements

09:40 - 10:40
Short talks
Oren Melamud (BIU)A Two Level Model for Context Sensitive Inference Rules
Tal Baumel (BGU)Query-Chain Focused Summarization
Abel Browarnik (TAU)Ontology Learning-
departing from the ontology layer cake

10:40 - 10:55
Coffee Break

10:55 - 11:55
Keynote talk
Dr. Yoelle Maarek (Yahoo! Labs) Queries and Questions in Yahoo! Answers

11:55 - 13:30
Short talks
Esther Goldbraich IBMLeveraging NLP to understand deviations from Clinical Practice Guidelines
Amnon LotanBIUTruthTeller: Annotating Predicate Truth
Oren TsurHUJIDon’t Let Me Be #Misunderstood:
Predicting User Word Preferences
According to Cognitive, Physical and Domain Constraints
Fiana RaiberTECHContent-Based Relevance Estimation on the Web Using Inter-Document Similarities
Effi LeviNITENiteRater: a Framework for Automated Scoring of Essays in Hebrew

13:30 - 14:45
Lunch + Poster (Lobby)

14:45 - 15:45
Keynote talk
Dr. Guy Hoffman (IDC) Human-Robot Interaction and Collaboration

15:45 - 16:10
Coffee Break

16:10 - 17:10
Keynote talk
Dr. Fadi Biadsy (Google Research NY) GOOGLE’S VOICE SEARCH - FOCUSING ON ARABIC AND HEBREW

17:10 - 17:15
Closing Remarks

17:15
End of ISCOL


Abstracts

Dr. Yoelle Maarek from Yahoo! Labs will talk about "Queries and Questions in Yahoo! Answers":

With the expansion of Web search engines, users have learned to stop asking proper natural-language questions (e.g., "What is the height of Corcovado?"), in favor of issuing queries (e.g., "height of Corcovado"). This transition has been driven by users becoming aware that search engines would not "understand" their questions, but rather attempt at locating Web pages that match, perfectly or approximately, their query phrase. Queries have somehow developed their own characteristics and have been shown to follow their own language model. Still, users do not see their needs always satisfied by search engines. This happens in three cases, either (1) the search engine did not succeed in properly accessing, indexing or retrieving relevant content, or (2) the issued query did not adequately represent the users' needs, or, finally, (3) relevant content does not exist yet for this specific need. Community-question answering sites have been devised to precisely address this third issue, as they allow users to turn to other users in order to get their needs addressed. In the latter case, searchers revert to askers: as they communicate with human beings, rather than machines, they become more verbose and express their needs as real questions. In this talk, we discuss the evolving relationship between questions and queries on the Web, how they differ and complement each other, and how users or machines have gone back and forth between these two forms of expressing a user's need. We show, in a typical application of big data analysis, how query logs, question logs and more interestingly query-to-question logs can be mined to bring new insights, and support new features to assist users in their quest for information.

Dr. Guy Hoffman from IDC:
Title: Human-Robot Interaction and Collaboration
Within decades, robots are expected to enter homes, offices, schools, and nursing facilities, and we should expect them even sooner in operating rooms, battlefields, construction sites, and repair shops. Successful human-robot interaction design is key to enable these robots to work smoothly and intuitively with untrained human partners. This challenge has led to the emerging field of Human-Robot Interaction (HRI) research, investigating computational, cognitive, and design approaches to human-adaptive robots, and evaluating people's responses to such robots.
This talk gives an overview of seminal and current HRI research, as well as my own research in human-robot collaboration. I discuss the design of non-verbal behaviors such as turn-taking, gaze, and gestures for shared attention, which we have implemented on an expressive humanoid robot, and how it impacts joint task performance between a human and a robot. In another project, we evaluate the role of anticipation of the human's actions, showing effects on team performance, as well as on the human partner's sense of the robot's trustworthiness and commitment. I then describe an embodied cognitive framework for robots, based on a model of perceptual simulation and cross-modal reinforcement, demonstrating how this model supports physical practice in an ensemble, and its effects on human subjects. For example, we see effects on the perception of the robot's intelligence, credit, and even gender, but also the induction of stress and self-deprecation.
In the field of entertainment robotics, I present a robotic theater control system using insights from acting theory, which enables robotic nonverbal behavior that is both reactive and expressive. I then discuss an interactive robotic Jazz improvisation system that uses embodied gestures for musical expression, enabling simultaneous, yet responsive, joint improvisation.
Finally, I present the design of a smartphone-based media companion robot. Human-subject studies show effects on music enjoyment and social presence when the robot responds to music that participants listen to, but no apparent sensitivity to the beat-alignment of the robot's motion.

Dr. Fadi Biadsy from Google Research New York:
Title: GOOGLE’S VOICE SEARCH - FOCUSING ON ARABIC AND HEBREW

Google voice search has changed the way users interact with their devices. Users speak their search queries and questions, typically using a mobile phone, and the system returns transcriptions, answers, and web search results. In this talk, I will describe Google’s voice search system. I will focus on the automatic speech recognition problems specific to Arabic and Hebrew, such as lack of diacritization/nikud and support for multiple Arabic dialects. We will compare recognizers for ?ve Arabic dialects with the potential to reach more than 125 million people in Egypt, Jordan, Lebanon, Saudi Arabia, the United Arab Emirates (UAE), and 7 million people in Israel. We compare systems built on diacritized vs. non-diacritized text, and show results of cross-dialect experiments.

Posters

Ofer BronsteinBIUTextual Entailment with Variables
Gitit KehatTAUStatistical transliteration of Judeo Arabic
Oleg RokhlenkoYahoo!Generating Synthetic Comparable Questionsfor News Articles
Asher SternBIUAn Application-Oriented Task and Dataset for Implicit Argument Recognition
Lili KotlermanBIUSentence Clustering via Projection over Term Clusters
Iddo AviramBGULDA with Redundancy
Iddo AviramBGUEffective Topic Modeling for Query Retreival
Avihai MejerYahoo!From Query to Question in One Click: Suggesting Synthetic Questions to Searchers
Naama Twitto–ShmuelHaifaImproving Statistical Machine Translation by Automatic Identifcation of Translationese
Nir OfekBGUUnsupervised Methodology for Context-Level Sentiment Analysis
Marina LitvakSCECross-lingual training of summarization systems using annotated corpora in a foreign language
Marina LitvakSCEMining the Gaps: Towards Polynomial Summarization
Chaya LiebeskindBIUStatistical Thesaurus Construction for a Morphologically Rich Language
Kfir BarTAUAutomatic Arabic Multiword Expression Boundary Detection
Dror MughazBIUEstimating the Birth and Death Years of Authors of Undated Documents using Undated Citationsþ
Raphael CohenBGUTowards Understanding of Medical Hebrew
Gabriel StanovskyBGUHebrew Paraphrase Recognition Using Deep Learning Architecture
Erel Segal HaleviBIUTextual Entailment - from Single Hypotheses to Grammar-Based Hypotheses Setsþ
Tomer HasidTAUStatistical Analysis of Orthographic Variationþ
Reut TsarfatyWeizmannA Unified Dependency Scheme for Automatic Morpho-Syntactic Analysisþ
Elena Even-SimkinBGUThe Application of Iconic and Systematic Character of 'Irregular'Forms for Computational Language Programsþ

Download - Papers / Presentations

Marina LitvakPolytope model for extractive summarization
Marina LitvakCross-lingual training of summarization systems using annotated corpora in a foreign language
Amnon LotanTruthTeller: Annotating Predicate Truth
Abel BrowarnikOntology Learning departing from the ontology layer cake
Naama Twitto–ShmuelImproving Statistical Machine Translation by Automatic Identifcation of Translationese
Raphael CohenTowards Understanding of Medical Hebrew
Iddo AviramEffective Topic Modeling for Query Retreival
Gabriel StanovskyHebrew Paraphrase Recognition Using Deep Learning Architecture
Oren MelamudA Two Level Model for Context Sensitive Inference Rules (paper,presentation)

Sponsored By: