This paper aims to describe a text-mining approach on a domain corpus (cork) within the theoretical framework of the dual dimension of terminology to create a terminological dictionary and correlate it with an ontology. We will make some considerations on (i) domain specificities; (ii) lexical markers; (iii) automatic corpus processing using Sketch Engine; (iv) representation of lexical networks using CmapTools; and (v) representation of the concept system using Protégé. The goal of the ontology is to logically support the coherence and quality of the natural language definitions contained in the terminological resource.
This paper aims to demonstrate a method for developing a terminological dictionary based on a domain ontology. To this end, we will describe the methods used to capture specialised lexical and conceptual knowledge from the corpus and use it to develop a dedicated ontology. The terminological resource will consist of a linguistic description of the specialised concepts, based on the formal definitions of the concepts that make up the cork ontology, the OntoCork [11].
The method used in this paper is corpus-driven. The corpus was compiled based on rigorous criteria
specific to terminological work [10], where the specialised context of text production is a key-element.
In this sense, the corpus is composed of technical explanatory and normative (standards) texts. For
corpus analysis, we used Sketch Engine2 to find and systematise lexical-semantic relationships. During
the corpus analysis process, we found two types of relevant knowledge-rich information [
Our method encompasses two stages: (i) From the linguistic analysis of the lexical markers, and the corresponding lexical-semantic relations observed between the terms, we systematise the results into lexical maps using CmapTools3.
(ii) Based on the previous stage, we proceed to the conceptual analysis and subsequent formal representation. The conceptual analysis grounds the identification of conceptual relations obtained by interpreting the lexical-semantic relations observed between two terms. To infer conceptual relations – such as the associative type – and to identify characteristics that will help us in the process of building concept systems, we resort to deductive mechanisms employing the Aristotelian formula: X=Y+DC to build OntoCork.
The Cork Corpus was built up from texts produced within the cork industry. The internal and
external criteria [
Degree of specialisation Source validation Type Content adequacy Synchronism (≤ 10 years)
Produced by experts and semi-experts Entities recognised as an authority Technical-explanatory; normative On cork/Cork stopper
Given the fast evolution of technology The corpus comprises 98 texts written in European Portuguese (see Figure 1). corpora collection academic articles
16%
These texts were produced by experts from different organisations and in different domains related to the cork industry. The texts were collected according to the following criteria: (1) texts produced by and for the scientific community in the domain of cork; (2) texts produced by experts for quasi-experts; and (3) texts produced by experts for non-experts.
Considering the 98 documents of the corpus, we have obtained the quantitative data shown in Table
2.
Tokens Words Sentences
1,712,652 1,217,968 48,031
For the corpus exploration and linguistic data analysis, we mainly focused on 43 texts produced in
two communicative settings, namely (i) expert–semi-expert, and (ii)
expert–quasiexperts/professionals, while the remaining 55 texts were used as a reference corpus [
Communicative setting of text production 16 29 4 43 16 27 n=98 Error! Reference source not found. 2: Subcorpus under focus (43 texts) based on the communicative setting of text production
Among the results presented in Table 2, the most frequent terms are “cortiça” [cork]4 and “rolha”
[stopper]. Given the high frequency of these two terms, we analysed the contexts in which they occur
in the subcorpus (43 texts) using the Word Sketch function as a first option, with which we identified
some candidate terms such as “ROLHA COLMATADA” [colmated stopper] (in capital letters). We then
moved on to simple queries (concordances) to search for polylexical terms containing adjectives in their
pattern. Once the most common morphosyntactic structures of terms were identified, we decided to
improve our search for terms and definitions employing advanced queries, namely through regex, so
that we could capture knowledge-rich contexts (KRC) [
Exploring the corpus with text mining methods
Based on the patterns we have identified within definitional contexts, we explored the subcorpus with advanced queries using regexes. For this paper, we will highlight two specific regexes that proved productive in isolating lexical relations between terms, but also in finding definitional contexts where the generic term is expanded in its syntax (see Table 3).
Definitional contexts (pt) (1) Rolha que foi submetida a um tratamento químico com o objectivo de desinfectar e/ou homogeneizar a cor e/ou branquear. (2) Rolha cuja superfície lateral foi submetida a uma operação de abrasão para a tornar cilíndrica ou diminuir o seu diâmetro.
Literal translation into English en: Stopper that was submitted to chemical treatment with the aim of disinfecting and/or homogenising the colour and / or bleaching en: Stopper whose side surface was submitted to an abrasion operation to make it cylindrical or to reduce its diameter.]
The first regex has the following structure: "rolha"[tag="V.P.*SF"], whose formulation aims to
match patterns as ONLY forms of “rolha” [stopper] followed by ANY past participle ONLY in the singular
and feminine inflection. For the elaboration of this regex, we considered the linguistic expressions used
repeatedly by the experts, such as the past participle co-occurring with a term (see Table 4). The
outcomes of this query, namely 69 hits, delivered the most productive patterns for identifying lexical
markers, such as “x foi submetida a y” [x was submitted to y], as well as terms whose morphosyntactic
structures fall under our search patterns, such as [Noun + Past Participle], e.g., “x acabada” [finished
X] or “X terminada” [finalised X] where X is a term and Y corresponds to a structure that has proved
4 Our translation
to be rich in knowledge information, i.e. information provided by the experts that allows us to perceive
their conceptualisations [
Considering the satisfactory results obtained, we decided to expand its formulation (regex 2): "rolha"[(tag="D.*"|(tag="S.*")]?[tag="A.*"]?"cortiça"?[]{0,4}"rolha"[]{0,4}[tag="V. P.*SF"]. In this case, we want to match a context in which the terms “rolha” [stopper] and “cortiça” [cork] may co-occur with either adjectives, past participles, or found duplicated, in addition to the functional forms. Out of the 55 hits matched by this regex, 48 were either a description or a definition.
From the whole set of descriptions or definitions semi-automatically extracted from the Cork Corpus, we decided to select ten (10) definitions for linguistic and conceptual analysis (see Table 4).
Ten (10) definitions to organise a typology of cork stoppers # 1 2 3 4 5 6 7 8 10 definitions (literal translations from pt)
10 definitions (pt) extracted from the Cork Corpus stopper rolha Product obtained from natural cork and / or Produto obtido da cortiça natural e / ou de cortiça agglomerated cork, consisting of one or more aglomerada, constituído por uma ou mais peças, pieces, intended to seal bottles or other containers destinado a vedar garrafas ou outros recipientes e a and to preserve their contents. (5.1 - NORM) preservar o seu conteúdo. (5.1 - NORM) STOPPER ROLHA piece of cork, usually cylindrical, conical or prismatic peça de cortiça, em geral cilíndrica, troncocónica ou quadrangular, sometimes with rounded or prismática quadrangular, por vezes de arestas chamfered lateral edges, consisting of one or laterais boleadas ou chanfradas, constituída por um several glued elements and intended to seal the ou vários elementos colados e destinada a vedar os containers or contribute to their water tightness. recipientes ou a contribuir para a sua (7.8 – TECH) estanquicidade (7.8 – TECH) natural cork stopper rolha de cortiça natural Stopper consisting entirely of natural cork Rolha totalmente constituída por cortiça natural. Note: Natural cork stoppers that have been Nota: As rolhas naturais que tenham sido submitted to the sealing operation (see 6.5.5) are submetidas à operação de colmatagem (ver 6.5.5) commonly referred to as colmated natural stoppers. são comummente designadas por rolhas naturais (5.5 – NORM) colmatadas. (5.5 – NORM) colmated natural cork stopper rolha de cortiça natural colmatada The colmated natural cork stopper is a stopper A rolha de cortiça natural colmatada é uma rolha made of natural cork in which its lenticels are filled feita de cortiça natural em que são obturadas as with a mixture of glues and cork powder from the suas lenticelas com uma mistura de colas e pó de dimensional finishing processes of natural cork cortiça proveniente dos acabamentos dimensionais stoppers. (6.1 – REP) das rolhas de cortiça natural. (6.1 – REP) agglomerated cork stopper rolha de cortiça aglomerada Stopper obtained by the agglutination of cork Rolha obtida pela aglutinação de granulado de granules with a size between 0,25 mm and 8 mm, cortiça com dimensão compreendida entre 0,25mm with addition of binders, by means of extrusion or e 8mm, com adição de ligantes, através de extrusão moulding and composed of at least 51% by weight ou moldagem e composta, pelo menos, por 51 % de of cork granules. (5.5 – NORM) granulado de cortiça, em peso. (5.5 – NORM) agglomerated stopper: rolha aglomerada: piece of agglomerated cork, obtained by extrusion peça de cortiça aglomerada, obtida por extrusão ou or moulding (3.1 – STUD) moldagem (3.1 – STUD) “SnoNnt+f.oB”ndp.id:sspitIksenoskrpstufhposoieersfrmdnd.aee(tsd5ui.gbr5nay–alactNioobOrnokR,dgM“ylnuo)”efidangdtogiclooamnteeesortarhtebeodntuchmoerbnkedarsn.d raedRNomioo“sllhnctbhaoa”oa:sdsnNfuoi+osetnrcsismlotitzasoaapddddoaeeossspc.i.ogo(nrr5tau.i5ççmaã–oncN,ao“OtrnupR”roMainld)cdeoicclaaodrotoinçsaúnmaugmeloroomudeeeramda technical stopper rolha técnica Technical stoppers are composed of a very dense As rolhas técnicas são constituídas por um corpo de body of agglomerated cork with disks of natural cortiça aglomerada, muito denso, com discos de cork glued to one end - or to both ends. Technical cortiça natural colados no seu topo – ou em ambos stoppers with one disk on each end are called 1+1 os topos. As rolhas técnicas com um disco em cada technical stoppers; those with two disks of natural topo são designadas rolhas técnicas 1+1. Com dois cork on each end are called 2+2 technical stopper; discos de cortiça natural em cada topo chamam-se and those with two disks glued at only one of the ends are called 2+0 technical stoppers. (6.1 – REP) rounded stopper Stopper whose edges of one or two ends were rounded by abrasion. (5.5 – NORM) marked stopper Stopper whose lateral surface or ends were marked in ink or by fire (7.6 – TECH) rolhas técnicas 2+2, e com dois discos em apenas um dos topos chamam-se rolhas técnicas 2+0. (6.1 – REP) rolha boleada Rolha cujas arestas de um ou dois topos foram arredondadas, por abrasão. (5.5 – NORM) ROLHA MARCADA Rolha cuja superfície lateral ou topos foram marcados a tinta ou a fogo. (7.6 – TECH)
For this paper, we will consider only one definition, namely <Rolha de cortiça natural> [natural cork stopper] (see line 3 in Table 4), to demonstrate our linguistic and conceptual analysis. However, instead of using the definitional statement written in Portuguese, we have decided to use its literal translation into English for clarity.
<Natural cork stopper>
stopper consisting entirely of natural cork Note: Natural cork stoppers that have been submitted to the sealing operation (see 6.5.5) are commonly referred to as colmated natural stoppers
In the second sentence – inserted as a note in the definition – another piece of information is obtained
from the analysis of the statement “natural cork stoppers that have been submitted to sealing operation”.
Here, the lexical marker is “submitted to” [submetidas à] and relates the term “natural cork stopper”
[rolha natural] to the term “sealing operation” [operação de colmatagem]. The term “sealing operation”
– which indicates an operation/activity – is related by the lexical marker “submitted to” [submetidas à]
to the term “natural cork stopper” – which we already know to be an object. The interpretation of their
meanings allows us to infer that the lexical-semantic relation established is meronymy, subtype
[ACTIVITY-FEATURE] [
The conceptual analysis corresponds to the second stage of the analysis of the definition in focus. The differential characteristics found in this definition are expressed by /natural cork/, /natural/, /colmated/ and /sealing operation/. The observations of this analysis are systematised in Table 6 and are based on the lexical markers found in the linguistic analysis of the definition. At the same time, based on the linguistic interpretation of the data, we extrapolated to conceptual relation identifiers.
As systematised in Table 6, we propose three conceptual relation identifiers, namely, (1) has_substance, (2) is_a, and (3) has_process.
(1) has_substance is expressed by the lexical marker “consisting entirely of” [totalmente
constituída por], which refers to the substance of the object. As we know from the linguistic analysis,
the term “natural cork” points to the notion of substance, a material that a given object can be made of.
Since <Stopper> is an object made of a substance, we propose the conceptual relation identifier
has_substance to represent such a semantic relation. This semantic relation mirrors a pragmatic
association - e.g., a thematic connection through virtue or experience, or a dependency between
concepts established by the proximity of time and space [
The dichotomy PRODUCT – RAW MATERIAL has twofold importance at this point of the conceptual
analysis: on the one hand, it underpins the subtype of the associative relation, while on the other hand,
it is included in the Aristotelian formula [
X=Y+DC natural cork
stopper [SPECIES] =
stopper [GENUS] + DC ? natural cork stopper [SPECIES] = stopper [GENUS] + natural cork
[DC] natural cork [GENUS] = cork [GENUS] + natural [DC] ? [SPECIES] = natural cork stopper [GENUS] +
sealing operation [DC]
colmated natural stopper [SPECIES] = natural cork stopper [GENUS] + colmated [DC]
Differential characteristi
cs /natural cork/ /natural/ /sealing operation/ /colmated/ the task of concept modelling, the characteristics stated in the definition under analysis. In order to use such a formula, one must first identify two concepts: the specific concept and its genus. [We will develop this further in the paper].
(2) is_a relation: <Natural Cork Stopper> is the subordinate concept, which we have labelled [SPECIES], and <Stopper> is the superordinate concept, which we have labelled [GENUS]. This assumption can be represented as: [natural cork stopper] SPECIES is_a [ stopper] GENUS. Once the genus and the species have been identified, we can then insert these two elements in the formula X SPECIES = Y GENUS + DC, where: X = natural cork stopper; Y = stopper. Differential characteristics are inferred in a second stage: considering that [stopper] PRODUCT has_substance [natural cork] RAW MATERIAL, we can conclude that X [natural cork stopper] = Y [stopper] + DC [natural cork]. The first statement of the definition conveys the information represented by the first interpretation above, with the dichotomy [SPECIES-GENUS], which can be represented in the form of a conceptual map (see Figure 4). Conceptual map 1 is built by applying a differentiae dichotomy in which the differential characteristic /natural cork/ underlies one of the subdivision criteria5.
Error! Reference source not found.4: Conceptual Map 1 - two composition types of <Natural_cork_stopper> in CmapTools Conceptual Map 1 (Figure 4) is the conceptual representation of the first statement of the definition, from which we have inferred that a <Natural cork stopper> is_a <Cork stopper>. Two axes of analysis are considered in this map: Substance and Parts (the ‘Parts’ axis was inferred from Definition 1; see Table 4). The conceptual information represented here, namely the axes of analysis Substance and Parts – whose underlying characteristics are /natural cork/, /mono piece/ and /multi piece/ – will be some of the coordinates for the elaboration of the formal description of the concept NaturalCorkStopper in 5 According to (ISO/FDIS 1087), the “subdivision criterion [is the] type of characteristic according to which a superordinate concept is divided into subordinated concepts.” (2019 (E), p. 5). Protégé. Finally, <Multi_piece_natural_cork_stopper> will help us to formally describe types of <Stoppers> composed of several Parts – not only made of <Natural_cork>, but also of <Agglomerated_cork> and <Mixed_cork>. Here, the characteristics fall under the axis of analysis ‘Parts’ and are the coordinates for modelling multi-part concepts.
(3) has_process: Following the same method, the analysis of the note from which we obtained the information: <Natural cork stopper> is submitted to /sealing operation/, was represented in a second map (Figure 5). This piece of information grounds the conceptual relation identifier we have named as has_process.
<Mono_piece_natural_cork_stopper_with_sealing_operation>
Conceptual Map 2 (Figure 5) is the representation of the two sentences of the definition in focus. Therefore, three axes of analysis are now considered: Substance, Parts, and Finishing Processes, to which the characteristics /with sealing operation/ and /without sealing operation/, were added. As represented in Conceptual Map 2, the characteristics /with sealing operation/ and /without sealing operation/ led us to a different level of concept representation, i.e., the concept <Mono_piece_natural_cork_stopper_with_sealing_operation>, verbally designated by “colmated cork stopper”, is a specialisation of <Mono_piece_natural_cork_stopper>, in turn, verbally designated by “natural cork stopper”. Therefore, these two concepts should not be treated at the same level, nor should they be defined in the same definitional context, either in natural language or in (semi)formal languages.
The conceptual relations we have inferred from the analysis of the lexical markers observed in the first five definitions (see Table 4), is summarised in Table 7.
‘is a’ ‘commonly referred as’ ‘is a’
Conceptual relation identifier is_a is_a is_a ‘intended to’
has_function PARTITIVE PARTITIVE stopper [SPECIES] = product [GENUS] + several pieces [PARTS=DC] stopper [SPECIES] = piece of cork [GENUS] + one element [PARTS=DC] stopper [SPECIES] = piece of cork [GENUS] + several elements [PARTS=DC]
As shown in Table 7, differential characteristics (DC) can be any characteristic in a given definition
according to the formula of an intensional definition [
For the task of building OntoCork, we used the editor Protégé [
Axis of analysis FUNCTION SUBSTANCE PARTS FINISHING PROCESS SHAPE
Format in Protégé hasFunction IsMadeOf hasStructure hasProcess hasShape
Type of conceptual relation associative relation, subtype [OBJECT-FUNCTION] associative relation, covering both subtypes [RAW MATERIAL – PRODUCT] and [MATTER/SUBSTANCE – PROPERTY] partitive relation [PART-WHOLE] associative relation, within the subtype [PROCESS-RESULT] associative relation, subtype [OBJECT-SHAPE]
For this paper, we will present the description of the characteristics that build up the formal definition of <Natural cork stopper>, a closure with a body-structure of 1 <Part> submitted to <Sealing process>, in addition to the classification provided by the reasoner HermiT 6 as a <Semi-finished> object (see Figure 6). 6 HermiT – a plugin reasoner of Protégé (http://www.hermit-reasoner.com/)
Figure 7 is the ontological representation of ColmatedMonoPieceNaturalCorkStopper, in Ontograf7, where we can observe several concepts systematised, either vertically: in a hierarchical dependency, or horizontally: in a pragmatic (associative) dependency, according to the differential characteristics. For clarity, we have decided to elide the visualisation of the associative relations between concepts that are not in focus in the following lines.
As illustrated in Figure 7, the ColmatedMonoPieceNaturalCorkStopper is a specification of MonoPieceNaturalCorkStopperWithFinishingProcess. The subsumption relation is represented by vertical blue arcs, and the associative relations are represented by horizontal dashed lines. The concepts 7 https://protegewiki.stanford.edu/wiki/OntoGraf ColmatedMonoPieceNaturalCorkStopper and LenticelsColmation are linked by the associative relation, subtype [PROCESS-RESULT]: hasLenticelsColmationOperation. This conceptual relation is based on the differential characteristic /with sealing operation/, which was drawn from the analysis of the definition of <Natural cork stopper>. Thus, hasLenticelsColmationOperation is the associative relation that induces the specification of MonoPieceNaturalCorkStopperWithFinishingProcess by differentia. Finally, it is also possible to see a hierarchical representation of FinishingProcesses, in which the involved operation of the concept we have just described is assigned as the most specific concept of this hierarchy. The interpretation of this subsumption is: LenticelsColmation is a kind of QualityTreatment, which is a kind of SurfaceTreatment, which in turn is a kind of SemifinishingProcess, all of these are kinds of FinishingProcesses.
With this research, we wanted to explain the method used to build an ontology from human analyses of linguistic data. Linguistic and conceptual levels of analysis are to be analysed in relation to one another but as distinct phenomena. Texts are vehicles for knowledge transfer. Analysing texts to extract the characteristics of concepts, linguistically expressed by lexical markers pointing to lexical-semantic relations, allowed us to effectively capture the conceptual relations that are specific to the domain through the formula X=Y+DC. As demonstrated in this study, we were able to propose a preliminary conceptual organisation of the subject field. We have bridged three main aspects in our study: (i) the classical aspects of the Aristotelian logic; (ii) the methodology of our terminological work – where characteristics play a fundamental role in the analysis or the drafting of intensional definitions; and (iii) the formal definitions, for which we have used Protégé and the inherent Web Ontology Language (OWL) [12] to formally describe the concepts of the domain in order to relate them via abstract syntaxes and thus achieve formal reasoning, as concepts are consistently defined in a ‘reason-able’ ontology.
In future work, we intend to model the conceptual and the linguistic information contained in the resources we have developed, namely the ontology, the corpus, and a glossary (in progress) developed with Lexonomy8, as linked data with the use of interoperable Linked Open Vocabularies9.
8 https://github.com/elexis-eu/lexonomy 9 https://lov.linkeddata.es/dataset/lov/