Automatic Generation of Semantic Features and Lexical Relations Using owl ontologies

Keywords: Semantic Feature Analysis, Ontology, OWL, Antonymy, Synonymy. 
1 Introduction 
The componential (semantic features) approach to meaning [1] [2] [3] views word 
meaning as a set of atomic meaning-holding components, called features. The metho-
dology by which these features are identified and extracted is called Semantic Feature 
Analysis (SFA). SFA has been fruitfully applied in various disciplines such as
linguistics (contrastive linguistic studies [4], and identifying lexical relations [2]), 
psycholinguistics (used as a model for understanding and studying conceptual repre-
sentations of meaning in the human mind [6]), and language learning and teaching 
(for vocabulary acquisition, and reading comprehension [5], [6] [7]).
Despite its numerous applications, SFA is not without its problems, one being a 
cumbersome and time consuming task which requires intensive human effort and 

---

16 
M. Al-Yahya et al. 
expertise. In this paper, we approach this problem by utilizing ontologies to automati-
cally perform SFA, and generate componential formulae. Once generated, these
formulae can be used to extract various lexical relations such as synonymy and
antonymy.
Ontologies are computational models for representing semantics for a specific do-
main. With advancements in web technologies and the increase in web based applica-
tions, we are seeing interest in developing ontologies whether general and upper level 
such as CYC [8] and SUMO [9], or domain specific such as OWL WordNet [10].
Since ontological models provide an explicit and formal semantic representation, 
using such ontologies as a resource for meaning derivation will simplify the task of 
SFA, and thus lexical relation extraction.
Work reported in the literature on extracting semantic features of words and identi-
fying various lexical relations is mostly based on analysis of large sets of textual
corpora, and identification of lexico-syntactic patterns for which a specific lexical 
relation holds. These patterns can either be observed manually, or derived computa-
tionally by using lists of words related with a specific lexical relation, and then
analyzing text corpora to find patterns. One of the earliest reported methods for lexi-
cal relation identification is that reported in [11]. Hearst [11] presents a method for 
automatic acquisition of the hyponymy relation between words by using manually and 
automatically generated lexico-syntactic patterns. Similarly, the work described in 
[12] uses lexico-syntactic patterns generated both manually and automatically to ex-
tract synonyms and antonyms from textual corpora. The Espresso system presented in 
[13], uses lexico-syntactic patterns for extracting semantic relations, with emphasis on 
generality of patterns usable across different corpora genre. The semantic relations 
extracted by Espresso are "is-a, part-of, succession, reaction, and production". The 
method described in [14] uses a score function composed of feature weights to classi-
fy pairs of words as antonyms. The features are extracted from contexts of word pairs, 
and include; Inversely Proportional Distance between pair words, lexico-syntactic 
patterns, and significance determined by a lexical co-occurrence network built for 
each word. Our approach is different from that reported in the literature. It is a novel 
approach which does not depend on lexico-syntactic patterns or large textual corpora, 
instead it is based on rules and logic inherent in OWL ontologies for extracting word 
features and determining the lexical relation that exists.
This paper is organized as follows. Section 2 presents background on semantic feature 
analysis and how it is performed and represented. Section 3 provides a brief back-
ground on OWL ontologies. Section 4 presents our prototype system, and the metho-
dology by which semantic features are extracted from OWL ontologies and used to 
compute lexical relations. Section 5 presents results of testing and evaluating
the system using a number of OWL ontologies. Section 6 concludes the paper and 
provides outlook for future work. 
2 Semantic Feature Analysis
Semantic feature analysis is based on the componential analysis theory of meaning 
representation [1]. Within this theory, a word's meaning is represented in terms of 
separable components (features), which if taken together constitute the componential