Resource acquisition via an unsupervised WSD system

Channel:
Microsoft Research
Subscribers:
344,000
Published on ● Video Link: https://www.youtube.com/watch?v=-fNmbv8dFUE


Duration: 26:44
44 views
0

In the field of WSD, it has been established that supervised systems (systems that rely on sense labeled data) perform significantly better than unsupervised WSD systems. Unfortunately, acquiring manual sense annotated data has proven to be a tedious expensive task. In this talk, I will show you how I use my unsupervised system, SALAAM, to acquire high quality sense annotated data to be used by supervised WSD systems for training. The approach could effectively reduces manual annotations by at least 40. In this portion of my talk I will characterize some crucial identifying features for discovering which automatically annotated data could be useful and which ones do require hand labeling. Another use for SALAAM lies in the ability to bootstrap resources for different languages. I will illustrate how I build an Arabic WordNet with a relatively high accuracy based on human ratings and judgments. The results obtained are very much in agreement with results reported by builders of the EuroWordNet.



Other Videos By Microsoft Research


2016-09-05A Systemic Approach to Appraisal: Identifying Opinion and Sentiment in Text
2016-09-05Structure and stability of complex networks.
2016-09-05Confronting Reality: Doing What Matters to Get Things Done Right
2016-09-05Recent Developments in Linguistics -- And in the UW Department of Linguistics
2016-09-05Dartmouth Digital Dorm          
2016-09-05A CLP Approach to Modelling Systems
2016-09-05Dynamosaics: Dynamic Mosaics with Non-Chronological Time
2016-09-05Automatic Failure Diagnosis in Large-Scale Systems
2016-09-05The Economics of Software Dependability
2016-09-05An Overview of Recent CMU Research on Model-Based Face Processing
2016-09-05Resource acquisition via an unsupervised WSD system
2016-09-05Fast Infinite-State Model Checking in Integer-Based Systems
2016-09-05WeΓÇÖre Friends, Right?: Inside KidsΓÇÖ Culture [1/4]
2016-09-05Accelerated Democracy: How technology might change voting
2016-09-05When Can Formal Methods Make a Real Difference?
2016-09-05Dynamic Semantics of Programming Languages and Applications to Testing
2016-09-05Error Detection using Shape Analysis with Local Reasoning
2016-09-05Type-Safe, Generative, Binding Macros
2016-09-05Social Network Analysis meets the Semantic Web: What FOAF Reveals About LiveJournal
2016-09-05Distributed Multi-robot Exploration and Mapping
2016-09-051.Vision: Extraordinary Computing Experiences & 2. Robots for the Masses: Fiction or Reality


Tags:
microsoft research