In this article we present an ongoing work for extracting conceptual information from specialized-domain texts. Concepts are forms of dividing the world in classes and they are the fundamental pieces for constructing ontologies. In this sense, ontology learning is the (semi-) automatic support for constructing an ontology. Input data are required for the ontology learning and this data are the basic source from which to learn the relevant concepts for a domain, their definitions as well the relations holding between them. With this necessity in mind, we propose here a methodology that takes into account the level of synthetic judgements and word relevance in a sentence in order to filter out and rank sentences. Sentences with high relevance and low level of synthetic judgements should have at least a predicative verb characteristic of analytical definitions for being good candidates.
Concepts are one of the most fundamental pieces
of the cognition: humans daily use concepts for
interacting with others and the world. According
to Smith (1988), concepts mirror the way that we
divide the world into classes, and much of what
we learn, communicate, and reason involves
relations among these classes. Additionally,
Currently, due to the accelerated growth of digital information on the Web and other media as well the urgent necessity of obtaining relevant information in a fast and efficient way from these huge text sources, automated methods or approaches have been developed. For instance, in Maedche and Staab (2004) define ontology learning as a number of complementary disciplines that feed on different types of unstructured and semi-structured data in order to support a semi-automatic ontology engineering process. In line with this, Cimiano (2006) describes various sub-processes for constructing an ontology from texts where the concept extraction is an important phase. So, the ontology learning needs input data from which to learn the relevant concepts for a given domain.
According to these ideas, in this paper we sketch a methodology for recognizing candidates to analytical definitional contexts, according to the work developed by Sierra et al. (2008). We organize our work as follows: in section 2 we present general information about analytical definitions and the automated extraction of conceptual information. In section 3 we describe the function of adjectives as modifiers of a noun as well the distinction among descriptive and relational adjectives and the relation of descriptive adjectives with synthetic judgements in an attributive form. In section 4 we summarize the methodology proposed. In section 5 we show some preliminary results. Finally, in section 6 we present the future work.
We consider as conceptual information the information expressed by specialized definitions, particularly in analytical definitions constituted by Genus Term and Differentia, following the criteria formulated by Smith (2004). In fact, this author considers that information expressed by these kinds of definitions is relevant to create ontologies based in lexical relations, specifically hyponymy/hypernymy and meronymy/holonymy relations. Smith argues that these relations, from a philosophical point of view, are basic and universal. 2.1
An analytical definition is a formula for describing a concept, denoted by a linguistic tag, in terms of a superordinate concept (Genus Term), and a differentia distinguishing the concept defined from others with the same Genus Term. For example, the next definition provides a description of the concept lightning conductor using one of the most common verbs (i.e., to be) for introducing a definition. In this case, the genus is the concept device while the differentia describes the function of the lightning conductor: [Lightning conductor Term] is a [device Genus Term] [that allows to protect the electrical systems against surges of atmospheric origin Differentia]. 2.2
Sierra et al., (2008) proposed a based-pattern method for extracting terms and definitions in Spanish. This relevant information is expressed in textual fragments called definitional contexts (or DCs) and are constituted by: a term, a definition, and linguistic or metalinguistic forms, such as verb phrases, typographical markers and/or pragmatic patterns, for example:
The primary energy, in general terms, is defined as an energetic resource that has not been affected for any transformation, with the exception of its extraction.
We can see here a DC sequence formed by the term primary energy, the definition that resource that… and the verb pattern is defined as, as well other characteristic units such as the pragmatic pattern in general terms and the typographical marker (bold font) that in this case emphasizes the presence of the term.
For achieving this objective, the authors employ verb patterns operating as connectors between terms and definitions. Such patterns syntactically are predicative phrases (or PrP), configured around a verb that operates as a head of this PrP (e.g., to be, to characterize, to conceive, to consider, to describe, to define, to understand, to know, to refer, to denominate, to call, to name).
Based on Demonte (1999), adjectives are syntactic units modifying the noun’s meaning and associating it with one or various attributes. There are two kinds of adjectives which assign properties to nouns: descriptive and relational adjectives. On the one hand, descriptive adjectives refer to constitutive features of the modified noun. These features are exhibited or characterized by means of a single physical property: color, form, character, predisposition, sound, and so on: la silla verde (e.g., the green chair). On the other hand, relational adjectives assign a set of properties, i.e., all the characteristics jointly defining names as sea: puerto marítimo (e.g., maritime port). In terminology, relational adjectives represent an important element for building specialized terms, e.g.: inguinal hernia, venereal disease, psychological disorder and others are considered terms in medicine. In contrast, rare hernia, serious disease and critical disorder seem more descriptive judgments and closely related with a specific context. 3.1
In line with what was just mentioned, if we consider the internal structure of adjectives, two kinds of adjectives can be identified: permanent and episodic adjectives (Demonte, 1999). The first kinds of adjectives represent stable situations, permanent properties characterizing individuals. These adjectives are located outside of any spatial or temporal restriction (i.e., psicópata- psychopath). On the other hand, episodic adjectives refer to transient situations or properties implying change and with time-space limitations. Almost all descriptive adjectives derived of participles belong to this latter class as well all adjectival participles (i.e., harto-jaded, limpio-clean). Spanish is one of the few languages that in syntax represent this difference in the meaning of adjectives. In many languages this difference is only recognizable through interpretation. In Spanish, individual properties can be predicated with the verb ser, and episodic properties with the verb estar.
Another linguistic heuristics for identifying descriptive adjectives is that only these kinds of adjectives accept degree adverbs, and they can be part of comparative constructions, for example, muy alto (Eng.: very high). Finally, only descriptive adjectives can precede a noun because in Spanish relational adjectives are always postposed, i.e.: la antigua casa (Eng.: the old house). 3.2
According to Kant (2013), analytic sentences are those whose truth seems to be knowable by knowing the meanings of the constituent words alone (e.g., gynecologists are doctors), unlike the more usual synthetic ones (e.g., gynecologists are rich), whose truth is knowable by both knowing the meaning of the words and something about the world.
We believe that synthetic judgements in an attributive position (e.g., rich gynecologists) are common in non-relevant sentences in specialized domains. This kind of judgements can be recognized from the descriptive adjectives obtained by linguistic heuristics mentioned in section 3.1.
We present here our methodology for extracting
conceptual information from a medical domain
corpus. The input data consist of a corpus with
POS tagged with FreeLing
The heuristics assumed here in order to segment our corpus by sentences take into account that a sentence must be separated by a point, to have at least a main verb, and the number of words must be greater than 10 words because the most short DC would have a single word term, the most long predicative verb-is defined as, a possible article preceding genus, genus term and, in this case, some arbitrary limit of words for the differentia). 4.2
The set of sentences obtained by the above step are filtered out by considering predicative verbs mentioned in section 2.2, that is, if there is at least a predicative verb; then it is a good candidate to DC. For the case of to be, if it is the first word of the sentence, then it is discarded. 4.3
We have used the library of Natural Language
NLTK
In this work, we propose a phase of quantification of synthetic judgments in candidate sentences as a further filter of non-relevant sentences. We assumed here that synthetic judgments are descriptive adjectives in an attributive position (e.g., rare syndrome). So, the higher amount of synthetic judgments in a sentence, the more likely sentence is non-relevant. We considered the set of descriptive adjectives obtained by heuristics as a mechanism for this quantification of syntheticity.
Acosta, Aguilar and Sierra (2013) point out relational adjectives have a higher probability of being part of terms. The heuristics considered in this experiment are: <RG><AQ> <VAE><AQ> <D.*|P.*|F.*|S.*><AQ><NC> Where RG, AQ and VAE as tagged with FreeLing, correspond to adverbs, adjectives and the verb estar, respectively. The tags <D.*|P.*|F.*|S.*> correspond to determinants, pronouns, punctuation signs and prepositions. The expression <D.*|P.*|F.*|S.*> is a restriction to reduce noise, since elements wrongly tagged by FreeLing as adjectives are extracted without this restriction. 4.4
We evaluated relevance of simple words by
means of a corpus comparison approach by
applying the relative frequency ratio
The ranking of sentences is done by adding up the individual ranks of words present in the sentence. Formally, if s (that is, a sentence) has a length of n words, w1 w2 …wn, where n>10, then the ranking of the candidate s is the sum of the weights of all the individual words wi W, where W are all of the relevant words weighted as mentioned in section 4.4. In contrast, if wi W, then its weight is zero.
Considering descriptive adjectives automatically extracted by heuristics for quantifying syntheticity, the first results show to be a good filter in order to remove non-relevant fragments by setting thresholds related with the number of descriptive adjectives in sentences. At the same time, the ranking of words achieves to sort sentences according to its relevance for the domain. Additionally, given that only sentences with predicative verbs are considered, a subset of the better ranked sentences are analytical DCs. If we take into account words where relative frequency in reference is greater or equal than in domain (given its higher occurrence in reference than in domain, we assume they are non-relevant words) as part of this list for removing nonrelevant sentences by setting thresholds (here, nouns and adjectives are included) improve significantly the results.
In a future phase of this experiment, we will implement a syntactic phase in order to remove more non-relevant sentences. For instance, sentences with to be verb are the most common sentences and which produce so much noise in results. Given this, we consider that a syntactic phase capable to assure the occurrence of specific syntactic structures will be an important advance in order to perform a better filtering. On the other hand, we will continue with the recollection of more information for increasing the sections of science and technology in our reference corpus, in order to improve the word weighing and the calculation of relevance sentences.
This paper has been supported by the National Commission for Scientific and Technological Research (CONICYT) of Chile, Project Numbers: 3140332 and 11130565. semantic relation extraction. Terminology, 14(1): 74-98.
Barry Smith. 2004. Beyond concepts: ontology as reality representation. In Formal Ontologies in Information Systems, ed. by Achille Varzi and Laure Vieu, pp. 73-84., IOS Press, Amsterdam.
Edward Smith. 1988. Concepts and Thought. In Psychology of human thought, ed. by Robert J. Sternberg, pp. 19-49. Cambridge University Press, Cambridge, UK.
Jorge Vivaldi. 2004. Extracción de candidatos a términos mediante la combinación de estrategias heterogéneas. Ph. D. Dissertation. IULA -UPF, Barcelona.