Loading...
Please wait, while we are loading the content...
Similar Documents
5. Sussna M. Word Sense Disambiguation for Free-text Indexing Using a Massive Semantic Network. Proceedings 5.2 Disambiguating Ambiguity
| Content Provider | Semantic Scholar |
|---|---|
| Author | Sebastiani, Fabrizio Cawsey, Alison Rijsbergen, Keith Van Krovetz, Bob Campbell, Iain |
| Copyright Year | 1997 |
| Abstract | process would adversely affect the retrieval performance. However, the results presented here suggest that this introduced ambiguity may not, in fact, be such a problem. Overall, the results presented in this paper appear to confirm 'common sense' beliefs, such as the ability of collocation to resolve word sense ambiguity and the high accuracy required of a disambiguator, with perhaps a little surprise as to the degree to which IR systems are resilient to ambiguity. It is hoped that this refining of the general appreciation of word sense ambiguity may be useful in identifying which areas justify further investigation within the context of IR. 5. Lesk M. Automatic sense disambiguation: how to tell a pine cone from an ice cream cone. occurring in a retrieved document) plays an important role in the impact of sense ambiguity to IR. This concurs with the findings of Krovetz and Croft. Intuitively this is perhaps not too surprising, after all, if a document is retrieved by matching on the query words: 'mammal', 'flying', 'vampire' & 'bat', it is unlikely that this particular use of 'bat' refers to the sporting implement. The final set of experiments investigated the effects on performance of a pseudo-word disambiguator operating at varying levels of accuracy. For these experiments ambiguity was introduced into the collection using size five pseudo-words. This additionally ambiguous collection was then disambiguated, but with a controlled amount of error. A retrieval was then run on the 'erroneously disambiguated' collection. This experiment was performed a number of times, each time with the percentage of correct disambiguations set to a different value. The results of two of these experiments are shown in Figure 7. As can be seen from the graph, disambiguation accuracy has a dramatic effect on performance. When the introduced ambiguity is disambiguated with an accuracy of 75%, the retrieval performance is actually worse than performance using the ambiguous collection. With disambiguation at 90% accuracy, performance is similar to that of the ambiguous collection, although a small improvement can be seen for retrievals based on queries composed of one or two words. There is some anecdotal evidence to suggest that in general, tools built for computational linguistics tasks need to operate at, at least 90% accuracy before they are of practical use. 6 Conclusions Using the novel experimental technique of introducing and removing ambiguity into a test collection in a controlled manner, insights into the significance of … |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://dis.shef.ac.uk/mark/cv/publications/papers/my_papers/SIGIR94.ps.gz |
| Alternate Webpage(s) | http://ir.dcs.gla.ac.uk/papers/Pdf/sanderson94b.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
