Amit Bagga and Alan Biermann. Analyzing the performance of message understanding systems. Technical Report CS-1997-01, Dept. of Computer Science, Duke University, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....has appointed GG as. president . 31 4.4 Event[2] LF for He succeeds. Talbot . 31 4.5 Event[3] LF for . Talbot, who resigned . 32 4.6 Event[2] Modified . 32 4. 7 Event[4]: Inferred Event . 33 8.1 Seed pattern set for Management Succession . 86 8.2 Seed pattern set for Mergers and Acquisitions . 90 8.3 Seed pattern set for Corporate Lawsuits . 91 ....

....selects which of the resultant LFs are to be output 1 , and formats them into the output structure specified by the user, e.g. into a database table. Slot Value class Leave Job company entity =# Euro.Info.Serv. position entity =# President person entity =# Talbot Table 4. 7: Event[4]: Inferred Event A thorough example of extraction which illustrates the operation of the various phases of Proteus is given in [19] 1 Some LFs are internal, such as Event 2. They are used for intermediate inference, and are not presented on output. The form of the output is given by the task ....

[Article contains additional citation context not shown here]

Amit Bagga and Alan Biermann. Analyzing the performance of message understanding systems. Technical Report CS-1997-01, Dept. of Computer Science, Duke University, 1997.

....a piece of text containing it. I have then used this to define a measure, the Domain Number which can be used to classify different domains with respect to IE tasks [Bagga 1997c] I have then used this classification mechanism to analyze the performance of various message understanding systems [Bagga 1997d] Currently, I am working on the University of Pennsylvania s system which will be submitted to the Seventh Message Understanding Conference (MUC 7) 1 The University of Pennsylvania s system will not take part in the full blown IE task for MUC 7; rather, it will take part in the coreference ....

Bagga, Amit. Analyzing the Performance of Message Understanding Systems, In Proceedings of the Natural Language Processing Pacific Rim Symposium (NLPRS '97), pp. 637-640, December 1997.

....done, we use task specific patterns to identify the facts of interest. 1 A process which may offer economic benefits when compared to purely manual extraction, in the same way that machine translation followed by post editing may be more efficient than manual translation. 2 Bagga and Biermann [7] discuss the relation between the success rate of extraction systems and the locality of the information in the text, measured in terms of the number of syntactic relations involved. The integration phase examines and combines facts from the entire document or discourse. It resolves relations of ....

Amit Bagga and Alan Biermann. Analyzing the performance of message understanding systems. Technical Report CS-1997-01, Dept. of Computer Science, Duke University, 1997.

No context found.

Amit Bagga and Alan W. Biermann. 1998. Analyzing the Performance of Message Understanding Systems. Journal of Computational Linguis- tics and Chinese Language Processing, 3(1):1--26, February.

No context found.

A. Bagga, and A. W. Biermann. 1998. Analyzing the Performance of Message Understanding Systems.

....Science Duke University Durham, North Carolina 27708 0129 February 20, Analyzing the Performance of Message Understanding Systems Based On Coreferencing Amit Bagga Dept. of Computer Science Duke University Durham, N. C. 27708 0129 amit cs.duke.edu February 20, 1 Introduction Earlier, in [Bagga 1997], we had described a method of classifying facts (information) into categories or levels; where each level signifies a different degree of difficulty of extracting the fact from a piece of text containing it. We then used this method to analyze the performance of three Message Understanding ....

....domain, the performances of the three systems take a hit on facts that contain a large number of coreferences. But, unlike MUC6, the hit taken by the three systems is not the same. For example, SRI s system takes a much bigger hit on facts that contain one coreferent node than NYU s system. In [Bagga 1997], we had observed that NYU s system does better on higher level facts than the other two systems. This observation can be explained by the performance of NYU s system on facts that contained one or more coreferent nodes. From Figure 14 we can observe that NYU s system performed better on such ....

[Article contains additional citation context not shown here]

Bagga, Amit, and Alan W. Biermann. 1997. Analyzing the Performance of Message Understanding Systems, Department of Computer Science, Duke University, Techreport CS-1997-01, January 1997.

....domain. Our analysis shows that all the three systems consistently did much worse on higher level facts. In addition to confirming Hirschman s hypothesis, the analysis also shows that higher level facts are indeed harder to extract. Some details of the analysis are given later in this paper. (Bagga 1997) gives the complete details about the analysis. Building the Semantic Networks As mentioned earlier, the level of a fact for a piece of text depends on the semantic network constructed for the text. Since there is no unique semantic network corresponding to a piece of text, care has to be taken ....

....module. As mentioned earlier, a level k fact is a union of k level 1 facts. Therefore, when retrieving such a fact, a system has to identify each of the k components and then the coreferencing module has to piece these k facts together. More details on such an analysis can be found in (Bagga 1997). Conclusion The level of a fact with respect to a semantic network for a piece of text provides a new method of classifying a fact based on the degree of difficulty of extracting it from that text. The analysis of the degree of difficulty of understanding a text in a domain comes as a byproduct ....

Bagga, Amit. 1997. Analyzing the Performance of Message Understanding Systems, Submitted to the Fifth Workshop on Very Large Corpora, (WVLC-5), 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC