Brought to you by EarthWeb
IT Library Logo

Click Here!
Click Here!

Search the site:
 
EXPERT SEARCH -----
Programming Languages
Databases
Security
Web Services
Network Services
Middleware
Components
Operating Systems
User Interfaces
Groupware & Collaboration
Content Management
Productivity Applications
Hardware
Fun & Games

EarthWeb Direct EarthWeb Direct Fatbrain Auctions Support Source Answers

EarthWeb sites
Crossnodes
Datamation
Developer.com
DICE
EarthWeb.com
EarthWeb Direct
ERP Hub
Gamelan
GoCertify.com
HTMLGoodies
Intranet Journal
IT Knowledge
IT Library
JavaGoodies
JARS
JavaScripts.com
open source IT
RoadCoders
Y2K Info

Previous Table of Contents Next


3.5. SEMANTIC ANALYSIS

Having determined the structural properties of an NL statement, sentence, or utterance, thanks to the use of the morphological and syntactic tools, we should now utilize the final syntactic structure and the semantic categories associated with the individual words to construct the overall "meaning" of the sentence. It is more or less admitted that, in its more general form, this meaning should consist of a formal statement expressed by making use of some "knowledge representation language" in the Artificial Intelligence acceptation of this term.

In reality, both the aims and the proper methods of the semantic phase are not very well defined, and this phase is much more fuzzy and problematic than the previous syntactic phase. The central problem concerns, of course, the fact that it is very difficult to find agreement on what the "meaning" of a statement can be -- and this, of course, has an heavy influence on the choice of the final representation and on its degree of "deepness." In this context, the traditional assumption of theoretical computational semantics consists of assuming that the determination of the meaning of a sentence can be equated with the elucidation if its "truth conditions," i.e., determining what the world would be like if the sentence were true. Apart from the problems that can rise when we try to adapt this sort of definition to questions and imperatives, it is evident that, when the aim is that of building up practical NLP components of KBSs, the main interest is certainly not in the problematics concerning true or false sentences. Intuitively, what we would like to obtain is a sort of standardised representation, independent from the different natural languages, unambiguous, and characterized by a well-defined set of basic inferences, that could be easily interfaced with, e.g., a DB language like SQL, the formal language of a robot command system, etc. Unfortunately, theoretical semantics is of little help here.

Moreover, apart from this type of epistemological consideration, there are all sorts of practical problems that make semantic analysis so difficult. For example, even if an agreement could be reached about the formal representation of the meaning, it must be considered that the meaning of a written or spoken statement is always sensitive to the context of the discourse: the context can be used to determine the proper signification of a word, the signification of the overall statement, or to determine how the statement can be used. Formalization of the context is a very advanced research topic where few concrete results have been already obtained -- even if we can admit that, e.g., dealing automatically with the anaphora problems according to the modalities mentioned, before, in Section 3.4.4.2, is already a first way of dealing automatically with the context problems. This is why the meaning representations concretely used are still, in practice, largely independent from the context, and based on the implicit assumption that the context-independent part of this meaning is, nevertheless, worthwhile to be represented, and sufficient for allowing a variety of useful applications. Of course, this is only an approximation.

The reasons expounded before are only a part of those that can be used to explain why: (1) whereas there are now a significant number of systems that display a (relatively) wide syntactic coverage, there are still few that can provide a corresponding degree of semantic coverage; (2) for the last, the label "semantic analysis" is used to identify procedures that have very few characteristics in common. Very roughly, we can distinguish three classes of procedures making use of some sort of "semantics": (1) procedures that are still, basically, of a syntactic nature, and where the semantic component is used, e.g., to solve some of the parsing ambiguities examined (see Section 3.4.4); (2) procedures characterized by some form of shallow semantic analysis; (3) procedures aiming to produce a sort of (quite) complete, formal representation of the "meaning" of the original statement. We shall now examine these three classes in turn.


Previous Table of Contents Next

footer nav
Use of this site is subject certain Terms & Conditions.
Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details.