The paper describes the application of Explanationbased Learning for efficient processing of constraintbased grammars. The idea is to generalize the derivations of training instances created by normal parsing automatically and to use these generalized deri
Application of Explanation-based Learning for E cient Processing of Constraint-based Grammars In Proc. of 10th IEEE on Arti cial Intelligence for Applications, San Antonio, Texas, 1994
Gunter Neumann, DFKI 66123 Saarbrucken, Germany neumann@dfki.uni-sb.de
Abstract The paper describes the application of Explanationbased Learning for e cient processing of constraintbased grammars. The idea is to generalize the derivations of training instances created by normal parsing automatically and to use these generalized derivations (called templates) during the run-time mode of the system. In case a template can be instantiated for a new input, no further grammatical analysis is necessary. The approach is not restricted to sentential level but can be applied also to arbitrary phrases. Therefore, the EBL method can be interleaved straightforwardly with normal processing to get back exibility that otherwise would be lost.
In the paradigm of feature-based grammar formalisms and theories, one line of research has been oriented towards developing high level languages which are based exclusively on constraint speci cations and which allow grammar writers to specify di erent levels of linguistic abstraction in one homogeneous data structure. In hpsg 4], currently one of the most prominent member of this class of formalisms the constraintbased view leads to a very schematic representation of universal and language speci c linguistic principles. From a computational point of view, a hpsg describes the search space of possible grammatical structures in a highly abstract and implicit characterization, so that phrasal rules have to be instantiated on-line using phrasal schemata and lexical information. However, on-line processing of such highly speci ed knowledge demands an enormous amount of computation, especially when declarative speci cations should be used in both parsing and generation. Currently, various lines of research are pursued to compensate for this de ciency and to obtain a better 1
1 Introduction
performance. Among these are compilation into lower level formats for which more e cient algorithms exist (e.g., 2], who investigates the compilation of hpsg into Tree Adjoining Grammar), or strategies which allow the attachment of control information determining the order of evaluation to constraints, e.g., 8]. In this paper we describe a di erent approach which is based on explanation-based learning. Rather than modifying the grammar sources or controlling the constraint evaluation process, this approach allows the automatic extraction of sublanguages from a competence base and its application within a natural language system. The basic advantages of this approach are: Processing of typical occurring grammatical constructions is vastly sped up using simple and efcient mechanisms, interleaving with normal processing allows for exibility that would otherwise be lost. The acquired structures are justi ed generalizations with respect to the competence gramm
ar in use. In the next section we discuss brie y the relationship between explanation-based learning and natural language processing. Our approach is described in detail in sections 3 to 5. In section 6 we relate our approach to other approaches, namely with that of 5, 6]. In section 7 we shortly outline some possible extensions of our approach.
Explanation-based learning (EBL) is a technique by which an intelligent system can learn by observing examples. An EBL system derives justi ed generalizations from training instances on the basis of declarative background knowledge. As a method, EBL
2 Explanation-based Learning and Natural Language Processing
On the Application of ... 16页 2下载券 A Habit-Based Explanat... ...RESOURCE-BASED EXPLANATION OF ENTRY MODE CHOICE Varinder M Sharma; M Krishna...
On the Application of ... 16页 2下载券 A Habit-Based Explanat... ...RESOURCE-BASED EXPLANATION OF ENTRY MODE CHOICE Varinder M Sharma; M Krishna...
热门文档