![]() |
|||
![]()
|
![]() |
![]() |
![]() |
3.5.2. Shallow Semantic Representations We will first examine a very well-known semantic tool, i.e., the "semantic grammars," and we will then look briefly at some of the semantic techniques used in the context of the MUC conferences (see subsection 2.4). 3.5.2.1. Semantic Grammars The basic idea behind the use of semantic grammars is that of unifying the two phases of syntactic and semantic analysis. This is obtained through the use of production (rewriting) rules characterized by the two following, main properties: (1) they produce parse trees where the categories of the grammar (i.e., the non-leaf nodes that appear in the tree, like NP or VP in the traditional syntactic analysis) do not coincide necessarily with the traditional syntactic categories; (2) these categories have been chosen to correspond as closely as possible to the desired inputs of the target program (e.g., a DB query program) and to the set of actions this last program must execute. In other words, the rules of a semantic grammar contain, hard-wired, the semantic information about the particular domain they are dealing with.
S --> QUERY SHIP-PROPERTY of SHIP QUERY --> what is | tell me SHIP-PROPERTY --> the SHIP-PROP | SHIP-PROP SHIP-PROP --> speed | length | type SHIP --> the SHIP-NAME | the fastest SHIP2 SHIP-NAME --> Kennedy | Kitty Hawk | Constellation | ... SHIP2 --> COUNTRYS SHIP3 | SHIP3 SHIP3 --> SHIPTYPE LOCATION | SHIPTYPE SHYPTYPE --> carrier | submarine | ... COUNTRYS --> American | British | Russian | ... LOCATION --> ... FIGURE 12 An example of semantic grammar. Figure 12 shows a (simplified) subset of the rules contained in one of the semantic grammars developed in the LADDER's context: LADDER was a famous NL interface to DBs developed in the late 1970s by G. Hendrix and colleagues, see, e.g., Androutsopoulos et al., 1995. It is very evident that, making use of this sort of grammars, we will produce -- when parsing, e.g., a query like "What is the speed of the Kennedy?" -- a parse tree having as intermediate nodes, in-between S and the actual words, very ad-hoc categories like QUERY, SHIP-PROPERTY, or SHIP. The main computational advantage of this type of tool is the possibility of avoiding the need of a special apparatus -- e.g., a system of variables and constraints on the variables -- in order to enforce the semantic constraints. The disadvantages are self-evident: since semantic grammars contain, hard-wired, the knowledge proper to a specific application domain, a new semantic grammar must be developed whenever the system must be ported to another domain. This is why this sort of tool has been progressively abandoned, even for applications, like the NL interfaces for DBs, where the semantic grammar approach was (formerly) particularly cherished. However, semantic grammars are still mentioned in a recent survey (Androutsopoulos et al., 1995); moreover, a revival of this sort of tools can now be detected in the Spoken Language Understanding systems domain (see Section 4).
|
![]() |
|
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. |