Yael Ravin

Disambiguation of Proper Names in Text

Identifying the occurrences of proper names in text and the entities they refer to can be a difficult task because of the many-to-many mapping between names and their referents. We analyze the types of ambiguity --- structural and semantic --- that make the discovery of proper names difficult in text, and describe the heuristics used to disambiguate names in Nominator, a fully-implemented module for proper name recognition developed at the IBM T. J. Watson Research Center.

Knowledge portals and the emerging digital knowledge workplace

A fundamental aspect of knowledge management is capturing knowledge and expertise created by knowledge workers as they go about their work and making it available to a larger community of colleagues. Technology can support these goals, and knowledge portals have emerged as a key tool for supporting knowledge work. Knowledge portals are single-point-access software systems intended to provide easy and timely access to information and to support communities of knowledge workers who share common goals. In this paper we discuss knowledge portal applications we have developed in collaboration with IBM Global Services, mainly for internal use by Global Services practitioners. We describe the role knowledge portals play in supporting knowledge work tasks and the component technologies embedded in portals, such as the gathering of distributed document information, indexing and text search, and categorization; and we discuss new functionality for future inclusion in knowledge portals. We share our experience deploying and maintaining portals. Finally, we describe how we view the future of knowledge portals in an expanding knowledge workplace that supports mobility, collaboration, and increasingly automated project workflow.

Is Hillary Rodham Clinton the president?: disambiguating names across documents

A number of research and software development groups have developed name identification technology, but few have addressed the issue of cross-document coreference, or identifying the same named entities across documents. In a collection of documents, where there are multiple discourse contexts, there exists a many-to-many correspondence between names and entities, making it a challenge to automatically map them correctly. Recently, Bagga and Baldwin proposed a method for determining whether two names refer to the same entity by measuring the similarity between the document contexts in which they appear. Inspired by their approach, we have revisited our current cross-document coreference heuristics that make relatively simple decisions based on matching strings and entity types. We have devised an improved and promising algorithm, which we discuss in this paper.

A Hybrid Approach to Natural Language Web Search

We describe a hybrid approach to improving search performance by providing a natural language front end to a traditional keyword-based search engine. The key component of the system is iterative query formulation and retrieval, in which one or more queries are automatically formulated from the users question, issued to the search engine, and the results accumulated to form the hit list. New queries are generated by relaxing previously-issued queries using transformation rules, applied in an order obtained by reinforcement learning. This statistical component is augmented by a knowledge-driven hub-page identifier that retrieves a hub-page for the most salient noun phrase in the question, if possible. Evaluation on an unseen test set over the www.ibm.com public website with 1.3 million webpages shows that both components make substantial contribution to improving search performance, achieving a combined 137% relative improvement in the number of questions correctly answered, compared to a baseline of keyword queries consisting of two noun phrases.

A TOOL FOR INVESTIGATING THE SYNONYMY RELATION IN A SENSE DISAMBIGUATED THESAURUS

This paper describes an exploration of the implicit synonymy relationship expressed by synonym lists in an on-line thesaurus. A series of automatic steps was taken to properly constrain this relationship. The resulting groupings of semantically related word senses are believed to constitute a useful tool for natural language processing and for work in lexicography.

Grammar errors and style weaknesses in a text-critiquing system

Grammar errors and style weaknesses identified by CRITIQUE, a text processing system developed at the IBM T.J. Watson Research Center, are discussed. Linguistic criteria for distinguishing between grammar and style are drawn first. These criteria are reflected in the messages issued by CRITIQUE to the user. Then, a computational criterion for distinguishing between grammar and style is discussed. This criterion is reflected in the implementation of the grammar-checking and style-checking mechanisms. Finally, it is explained how CRITIQUE operates when the criteria are in conflict: the implementation is driven by the computational criterion, while the display to the user remains faithful to the linguistic criteria. >

Who's who? Identifying concepts and entities across multiple documents

A number of research and software development groups have developed technology for identifying terms and names in documents and associating them with concepts and named entries, but few have addressed coreference of concepts and entities across multiple documents in a collection. Cross-document coreference is challenging, since a collection of documents consists of multiple discourse contexts, with a many-to-many correspondence between terms and names on one hand and the concepts and entities they refer to on the other. In this paper we describe extensions to our intra-document term and name identification for coreferencing concepts and entities across documents.

Parse Fitting and Prose Fixing

Processing syntactically ill-formed language is an important mission of a text-critiquing system. This chapter discusses how ill-formed input is treated by Epistle, the forerunner of Critique. Misspellings are highlighted by a standard spelling checker; syntactic errors are detected and corrections are suggested; and stylistic infelicities are called to the user’s attention. Central to the processing strategy is the technique of fitted parsing. When the rules of a conventional syntactic grammar are unable to produce a parse for an input string, this technique can be used to produce a reasonable approximate parse that can serve as input to the remaining stages of processing.