| Ohnishi Lab. |
|
Natural Language Processing Research Theme |
| Evaluating Abstract Qualification |
A paper abstract plays an important roles, because it gives the necessary information as to whether we should read it. This work describes a method for evaluating abstracts of technical papers. First, we define semantic roles of a sentence and give the classification of roles. Then, we investigate the relationships between surface information and each semantic roles of sentences in an abstract. Based on these relationships, we propose a method for inferring the semantic roles of sentences an abstract. We applied this method to 511 sentences in 77 abstracts and obtains the satisfactory results; the semantic roles of 413 sentences (about 81%) were correctly inferred. Finally, we evaluated an abstract by using a criterion that known information doesn't occupy greater part of an abstract and that the executions of the authors and their results are written. An a results are written. As a result, 8 of 50 abstracts didn't satisfy the criterion. |
