Abstract
TELIX is an OWL vocabulary to model text encoding and to facilitate
linguistic information exchange as RDF graphs. It extends SKOS XL, which allows
capturing lexical entities as RDF resources, and it overcomes SKOS expressive
limitations. TELIX introduces a number of classes and properties aiming to
provide natural language acquisition and extraction tools with powerful
mechanisms to build, on the one hand, interchangeable, multilinguistic lexical
resources (such as dictionaries or thesauri), and on the other hand, to
represent the outcomes of their text analysis (annotations content). The full
specification of TELIX can be found in [FIXME-Technical report ONTORULE-D1.4],
where complete details about the ontology and modeling decisions are
provided.
Status of This Document
This section describes the status of this document at the time of its
publication. Other documents may supersede this document. A list of current W3C
publications can be found in the W3C technical
reports index at http://www.w3.org/TR/.
This document is part of the TELIX Submission which comprises three
documents:
- TELIX Overview and Motivation
- TELIX Linguistic annotations
- TELIX Implementation and Use cases
Earlier versions of the TELIX specification have been published within the
wp activity of the ONTORULE project at TELIX
Annex
By publishing this document, W3C acknowledges that the Submitting Members have made a
formal Submission request to W3C for discussion. Publication of this document
by W3C indicates no endorsement of its content by W3C, nor that W3C has, is, or
will be allocating any resources to the issues addressed by it. This document
is not the product of a chartered W3C group, but is published as potential
input to the W3C Process. A
W3C Team Comment
has been published in conjunction with this Member Submission. Publication of
acknowledged Member Submissions at the W3C site is one of the benefits of W3C Membership.
Please consult the requirements associated with Member Submissions of section
3.3 of the W3C Patent Policy. Please consult the complete list of acknowledged W3C Member
Submissions.
1 Introduction
This section introduces TELIX (Text Encoding and Linguistic Information
eXchange), an OWL vocabulary for the representation linguistic information as
RDF graphs. TELIX is conceived as an extension of the W3C vocabulary SKOS XL.
TELIX introduces a number of classes and properties aiming to provide natural
language acquisition and extraction tools with powerful mechanisms to build, on
the one hand, interchangeable, multilinguistic lexical resources (such as
dictionaries or thesauri) in the web and to represent, on the other hand, the
outcomes of their text analysis.
Lexical-oriented web dictionaries and thesauri
TELIX is (partially) built on W3C SKOS eXtension for Labels (SKOS-XL), which
introduces term literals as RDF resources within SKOS conceptual schemes.
Moreover, TELIX fixes the interpretation of skosxl:Label as
lexemes, pairings of particular ortographic (and phonological) form with a
meaning representation. The meaning is supposed to be captured by a
skos:Concept or a domain-specific ontology concept. Canonical
forms of lexemes (lemmas) are the values of the property
skosxl:literalForm in a SKOS XL terminology.
As it was stated previously, TELIX does not distinguish between homonyms and
polysemous words, i.e. there are not words with multiple senses. A new lexeme
is created for each word sense and all the lexemes with the same citation form
are directly related via this property: homonym
Lexical relations
TELIX extends the property skosxl:labelRelation to capture the
different possible lexicographic relations between terms:
telix:synonym property, which captures a relation between two
different lexemes with the same meaning, e.g.: “car” and
“automobile”. This is a symmetric property among labels. The synonymy
is indirectly encoded by the usage of skosxl:prefLabel and
skosxl:altLabel. It is intended that terms used to document a
concept with these properties can be swapped when referring to it.
telix:homonym property, which captures a relation between two
terms with the same citation form, but with a different meaning. E.g.:
“bank” as financial institution and “bank” as “slope of land”.
It is a symmetric property among labels.
telix:antonym property, which captures a relation between
labels words which meanings are considered as opposite, e.g.: “tall” is
the antonym of “short”. It is symmetric property.
telix:hypernym property, which captures a relation between two
terms where the meaning of one term is considered broader (or more generic)
than the other, e.g., “vehicle” is a hypernym of “car”. This
property is the inverse of telix:hyponym.
telix:hyponym property, which captures a relation between two
terms where the meaning of one term is considered narrower (or more
specific) than the other, e.g., “car” is an hyponym of “vehicle”.
This property is the inverse of telix:hypernym.
Collocations
A collocation is
“any statistically significant co-occurence of words” [FIXME]. It is comprised
by a target word and its surrounding context (also known as local pattern).
A SKOS-XL label may have several patterns of appearance
in texts, thus TELIX introduces the concept telix:Context to
capture the idea of collocation.
The relation between the target label and its contexts is expressed via
the property telix:inContext. A telix:Context definition
considers the following aspects:
- The sequence of words surrounding the target label, which is expressed by means of the
datatype functional property
telix:collocation.
- The ±
k words to the right or left of the target, also known
as the window of the context, which is represented by means of the functional datatype property
telix:window.
- The probability of appearance of the collocation, captured by means of the
datatype functional property
probability.
- The link to the textual sources used to estimate the frequency of the
collocation, as local contexts are highly dependent of the collection
of documents. The property is called
telix:source and can be applied
to single documents or to a linguistic corpus. It is worth remarking that
the property telix:source is also functional, following the premise
that there is only one sense for each collocation in a given discourse
[21, 23].
Text Analysis
Text segmentation
TELIX proposes a global model to deal with documents which is twofold. On
the one hand, at one first level, by identifying the kind of resources TELIX
can deal with. On the other, by defining properly the textual units potentially
exploited by the model proposed by TELIX.
- Organization of documents: TELIX reuses the Dublin Core Type
Vocabulary to provide standard categories of content resources. Other kinds
of content resources, such as diagrams or videos are also representable
within the model. Moreover, resources can be organized in groups or
collections, for instance, a linguistic corpus.
- Structure of the document: This part of the data model deals
with the segmentation of the text structure. Documents are usually split
into different textual units which differ in scope and importance. TELIX
provides at least the following textual units:
Section,
Paragraph and Sentence. More refined textual
units, such as title, chapter, itemized list, etc., are possible, but are
out of the scope of the ontology. In any event, it is possible to extend
TELIX (or to combine with other vocabularies), like any other OWL ontology,
to fit the specific requirements of a particular application.
The Figure above shows the hierarchic minimal text segmentation provided by
TELIX, from Sentence, which is the minimal text segmentation and
part of the following level Paragraph, to Section the
upper level textual unit taken into account by TELIX.
Feature structures
Grammatical information such as the number (singular or plural) of a noun,
or a verb form (e.g. whether it is a present participle, an infinitive, etc.)
serve to describe lexical entries. Instead of associating words in the lexicon
with a single atomic category, TELIX treats a lexical category as a complex of
grammatical properties, using the notion standardly referred as feature
structure to model them. TELIX allows that any complex feature structure
regarding different grammar levels may be transformed to an equivalent RDF
graph.
It is out of the scope of this specification to provide a complete mapping
from features structures and types used in grammars such as HPSG or LFG, but
the TELIX model render it easily feasible.
Parts-of-speech, POS or lexical tags provide useful information about their
neighbors, i.e. the words with which they usually combine, are also
reperesented in TELIX by means of the mechanism of feature structures. To this
end, a class called PartOfSpeech draws together all the distinct
word classes and a new property, telix:pos
is intended to relate labels with their syntactic category. The latter must
be functional, as the same word can not be member of different word classes.
TELIX also bears a property enabling to express these hierarchies between
large lists of part-of-speech: telix:refines. This property
telix:refines, in conjunction with the TELIX part-of-speech tags, allows to
group large lists of elements, such as the one provided by C7 tagset, with 146
wordclasses, or the Brown corpus. This mechanism provides a certain, although
limited, degree of interoperability among different part-of-speech taxonomies,
as they may agree via the mediation of TELIX. Nevertheless,
telix:refines should not be mistaken as a subsumption mechanism.
Due to the functional restriction of telix:pos, a DL reasoner
would deduce that two part-of-speech tags related by means of
telix:refines are the same, which is not always desirable.
More details about this complex feature structure matter are given
in the TELIX Technical Report [FIXME].
Discourse analysis
Reference relations
In a given discourse, multiple expressions may refer to the same thing, or
in linguistic jargon, they share the same referent, i.e. they corefer. Natural
language expressions used to perform reference are called referring
expressions, which range from (1) definite/indefinite noun phrases: “the/a
dog in the street”; (2) proper nouns: “The Eiffel Tower” or
“Barcelona”; and (3) Pro-forms, such as pronouns and deictics: "she" or
"their", respectively.
TELIX introduces a new property called telix:corefers to
express coreference relationships between several refe[rring expressions within
the same text. To fully capture the semantics of correference,
telix:corefers is defined as symmetric and transitive. The
coreference is not restricted to a pro-form and its referent:
“[Charlie]i saw [his]i friends", but it is also
possible between independent referring expressions, such as: “[A
magpie]i has just taken us over. I love [that bird]i.
In addition, TELIX provides two inverse subproperties of
telix:corefers to refine the anaphoric relationship between
coreferent expressions.Notice that these properties work independently of the
linear sequence of the text. This means that anaphoras, which antecedents
appear prior in the discourse, and cataphoras, which antecedents come later,
are equally captured by these two properties.
telix:antecedent property, which relates a pro-form with the
referring expression it is bound to. telix:anaphora property, which links a referring expression
to its dependent pro-forms.
Discourse structure
Facing the discourse structure of a text, TELIX dissociates the discourse
analysis from the actual segmenation of the text, which provides a linear order
of a number of text spans. To this end, a new kind of entities are introduced
in the linguistic model, DiscourseUnit. These elements capture the
elementary components of a text discourse structure, which in fact may
correspond with a single (or a list of) textual units or other categories
coming from the syntactical analysis of the text (such as clause). An essential
requisite must be fulfilled: discourse unist can not overlaps in the text. The
property telix:spans relates the elementary discourse units with
the corresponding text segmentations.
TELIX also distinguishes between SimpleDiscourseUnit and
ComplexDiscourseUnit. The former captures the elementary discourse
meaning which correspond to non-overlapping fragments of text. The latter
represents the discourse composition from simpler units. Elementary and complex
discourse units come related with the property telix:composedBy
which allows to derive tree-like discourse structures in RDF.
On the other hand, TELIX provides a set of properties to represent the
rethorical relations between the discourse units, both elementary and complex
ones, following Marcu [FIXME] and Thomas and Mann [FIXME]. The flexibility of
RDF enables to express tree-like and graph-like discourse structures:
- Hypotactic properties, also known as mononuclear, where there
exists a dependence relation (nucleus-satellite) between the discourse
units.
- Paratactic properties, also known as multinuclear, which are
relations between units of equal importance (nucleus-nucleus).
5 Linguistic annotations
A linguistic annotation consists of any descriptive or analytic marks
dealing with raw language data extracted from texts or any other kind of
recordings. A large and heterogeneous number of linguistic features can be
involved. Typically linguistic annotations include part-of-speech tagging,
syntactic segmentation, morphological analysis, co-references marks, phonetic
segmentation, prosodic phrasing, intonation or gesture marks and discourse
structures, among others.
In business scenarios, there is an increasing need to interchange linguistic
information and annotations, as well as the source documents they refer to,
among software tools by different vendors. Text analysis and information
acquisition often require incremental steps with associated intermediate
results. Moreover, tools and organizations make use of shared resources such as
thesauri or annotated corpus. Therefore, there is a need of appropriate
standards that support this open information interchange. These standards must
provide means to model and serialize the information as files.
Some of the requirements for a linguistic annotation framework seems to be:
expressive adequacy, media independency, semantic adequacy, incrementality,
uniformity, openness, extensibility, human readability, processability
(explicitness) and consistency. The RDF framework features these properties,
and therefore it constitutes a solid foundation. RDF graphs make use of custom
vocabularies defined by ontologies. Therefore, we introduce TELIX, a
lightweight ontology that provides comprehensive coverage of linguistic
annotations and builds on previous resources, such as feature structures and
SKOS concept schemes. TELIX takes advantage from the RDF/OWL expressive power
while being compatible with previously developed materials such as lexicons and
terminolgies. Moreover, translations from traditional linguistic annotation
formats to RDF are possible.
6. TELIX at a glance
An alphabetical index of NAME terms, by class (concepts) and by property
(relationships, attributes), are given below. All the terms are hyperlinked
to their detailed description for quick reference.
Classes: AdjectivalPhrase, AdpositionalPhrase, AdverbialPhrase, Annotation, Aspect, CC, CD, Case, Category, ComplexDiscourseUnit, ConjunctionPhrase, Context, Corpus, DT, DiscourseUnit, EX, FW, Gender, GrammaticalFeature, IN, JJ, JJR, JJS, LS, LabelOccurrence, MD, Mood, NN, NNP, NNPS, NNS, NounPhrase, Number, PDT, POS, PRP, PRPS, Paragraph, PartOfSpeech, Person, Phrase, PostpositionPhrase, PrepositionalPhrase, RB, RBR, RBS, RP, SP, SYM, Section, Sentence, SimpleDiscourseUnit, TO, Tense, TextualUnit, Token, UH, VB, VBD, VBG, VBN, VBP, VBZ, VerbPhrase, Voice, WDT, WP, WPS, WRB,
Properties: abbreviationModifier, adjectivalComplement, adjectivalModifier, adverbialClauseModifier, adverbialModifier, agent, agreement, anaphora, antecedent, antithesis, antonym, appositionalModifier, argument, aspect, attributive, auxiliary, background, case, childNode, circumstance, clausalComplementWithExternalSubject, clausalComplementWithInternalSubject, clausalSubject, collocation, complement, complementArgument, complementizer, composedBy, concession, condition, conjunct, constrollingSubject, contrast, coordination, corefers, dependent, determiner, elaboration, elementOfCompoundNumber, enablement, evaluation, evidence, expletive, feature, follows, gender, head, homonym, hypernym, hyponym, hypotactic, inContext, infinitivalModifier, interpretation, justify, length, marker, modifier, mood, motivation, negationModifier, nominalSubject, nonVolitionalCause, nonVolitionalResult, nounCompoundModifier, number, numericModifier, object, occurs, offset, otherwise, paratactic, participialModifier, person, phrasalVerbParticle, pos, position, possessionModifier, possessiveModifier, precedes, prepositionalModifier, probability, purpose, purposeClauseModifier, realizes, referent, refines, relative, relativeClauseModifier, restatement, semanticDependent, sense, sequence, solutionhood, source, specifier, subject, summary, synonym, temporalModifier, tense, value, voice, volitionalCause, volitionalResult, window,
Instances: ablative, accusative, active, antipassive, applicative, causative, cc, cd, conditional, dative, distributivePlural, dt, dual, ex, femenine, first, future, fw, genitive, gnomic, imperative, imperfective, in, inchoative, indicative, instrumental, interrogative, jj, jjs, jussive, locative, ls, masculine, md, middle, neuter, nn, nnp, nnps, nns, nominative, optative, passive, past, paucal, pdt, perfective, plural, pose, potential, present, prp, prps, quadral, rb, rbr, rbs, rp, second, singular, subjunctive, sym, third, to, trial, uh, vb, vbd, vbg, vbn, vbp, vbz, vocative, wdt, wp, wps,
7. Cross-reference for TELIX terms
...
Class: telix:AdpositionalPhrase
URI: http://purl.org/telix#AdpositionalPhrase
Adpositional phrase - An adpositional phrase (category including both prepositional phrases and postpositional phrases) is a phrase containing an adposition in the head position and usually a complement such as a noun phrase.
- sub-class-of:
- telix:Phrase
[back to top]
Class: telix:Annotation
URI: http://purl.org/telix#Annotation
Annotation - Class annotation captures linguistic annotations as entities within the model. It describes the annotation itself and it is also used to identify which linguistic data belongs to a particular annotation or to another.
- sub-class-of:
- owl:Thing
- in-domain-of:
- telix:source
[back to top]
Class: telix:Context
URI: http://purl.org/telix#Context
Context - Also known as "local pattern", it is the word entourage that helps disambiguating the word itself.
- in-domain-of:
- telix:source
- telix:collocation
- telix:probability
- telix:window
- in-range-of:
- telix:inContext
[back to top]
Class: telix:Gender
URI: http://purl.org/telix#Gender
Gender - A linguistic feature showing a system in the grammar of some languages in which nouns. pronouns (and any modifier or other type of word affecting that noun, as adjectives or articles) are classified as belonging to a certain gender -often masculine, feminine or neuter.
- sub-class-of:
- telix:GrammaticalFeature
- in-range-of:
- telix:gender
[back to top]
Class: telix:GrammaticalFeature
URI: http://purl.org/telix#GrammaticalFeature
Grammatical feature - Grammatical features are elements into which linguistic units can be divisible in order to obtain better ways to analyse and describe a given language.
[back to top]
Class: telix:Mood
URI: http://purl.org/telix#Mood
Mood - A linguistic feature which describes the relation of the verb to reality or intent in speaking. It is different from grammatical tense or grammatical aspect, although these concepts are conflated to some degree in many languages. some of the most common moods include conditional, imperative, indicative, potential, subjunctive, etc.
- sub-class-of:
- telix:GrammaticalFeature
- in-range-of:
- telix:mood
[back to top]
Class: telix:Paragraph
URI: http://purl.org/telix#Paragraph
Paragraph - A distinct portion of a text with a particular idea, usually beginning with an indentation on a new line. Paragraphs are used to help the reader understand a full text by breaking down it.
- sub-class-of:
- telix:TextualUnit
[back to top]
Class: telix:Person
URI: http://purl.org/telix#Person
Person - A linguistic feature consisting on a deictic reference to a participant (speaker, addressee or others) in an event. This feature typically defies a language's set of personal pronouns and it also often affects verbs, nouns, pronouns, articles and possessive relationships.
- sub-class-of:
- telix:GrammaticalFeature
- in-range-of:
- telix:person
[back to top]
Class: telix:PostpositionPhrase
URI: http://purl.org/telix#PostpositionPhrase
Postposition phrase - A postpositional phrase is a phrase containing a preposition in the head (ending) position and usually a complement such as a noun phrase.
- sub-class-of:
- telix:AdpositionalPhrase
[back to top]
Class: telix:Sentence
URI: http://purl.org/telix#Sentence
Sentence - An expression in natural language, usually defined to indicate a grammatical unit, bearing minimal syntactic relation to the words that precede or follow it, often preceded and followed in speech by pauses, having one of a small number of characteristic intonation patterns, and typically expressing a statement, question, request or command.
- sub-class-of:
- telix:TextualUnit
- in-range-of:
- telix:occurs
[back to top]
Class: telix:Tense
URI: http://purl.org/telix#Tense
Tense - A linguistic feature that locates a situation in time, indicating when the situation takes place. Languages which have tense usually indicates it by a verb or modal verb, often combined with other features, such as aspect, mood or voice.
- sub-class-of:
- telix:GrammaticalFeature
- in-range-of:
- telix:tense
[back to top]
Class: telix:TextualUnit
URI: http://purl.org/telix#TextualUnit
Textual unit - The structure of a text can be segmented according to a number of criteria: paragraphs, sentences, chapters, titles, etc.
- sub-class-of:
- owl:Thing
- in-domain-of:
- telix:follows
- telix:precedes
- telix:length
- telix:offset
- telix:position
- telix:value
- in-range-of:
- telix:follows
- telix:precedes
[back to top]
Class: telix:Token
URI: http://purl.org/telix#Token
Token - A token is a contiguous string of alphabetic (sometimes also numerical) characters. Tokens are separated by whitespace characters, or by punctuation characters, as punctuation is not included in the resulting list of tokens. The class token provides an auxiliary entity to empower tools performing different interpretations of lexical entities, i.e., which are the materializations of terms in the text fragment (LabelOccurrence).
- sub-class-of:
- telix:TextualUnit
[back to top]
Class: telix:Voice
URI: http://purl.org/telix#Voice
Voice - A linguistic feature, also called diathesis, of a verb which describes the relationship between the action (or state) that the verb expresses and the participants identified by its arguments (subject, object). The most common voices are active and passive, although other kinds can be found along world languages, such as middle, causative, cooperative voices, etc.
- sub-class-of:
- telix:GrammaticalFeature
- in-range-of:
- telix:voice
[back to top]
Property: telix:phrasalVerbParticle
URI: http://purl.org/telix#phrasalVerbParticle
Phrasal verb particle -
- OWL Type:
- ObjectProperty
- sub-property-of:
- telix:modifier
- Domain:
- telix:LabelOccurrence
- Range:
- telix:LabelOccurrence
[back to top]
Property: telix:position
URI: http://purl.org/telix#position
Position - The position of a textual unit indicates the order of this textual unit in the family of sibling textual units within the context of its immediately superior textual unit.
- OWL Type:
- DatatypeProperty
- Domain:
- telix:TextualUnit
[back to top]
Property: telix:refines
URI: http://purl.org/telix#refines
Refines - Property refines enabling to express these hierarchies between large lists of part-of-speechs.
- OWL Type:
- ObjectProperty
[back to top]
Instance: telix:conditional
URI: http://purl.org/telix#conditional
Conditional - The conditional mood is used in the independent clause of a conditional sentence to refer to a hypothetical state of affairs, or an uncertain event that is contingent on another set of circumstances. This mood differs from the subjunctive mood, which occurs in dependent clauses.
- RDF Type:
- owl:Thing
- telix:Mood
[back to top]
Instance: telix:dual
URI: http://purl.org/telix#dual
Dual number - As in Slovene, a grammatical number that some languages use in addition to singular and plural. When a noun or pronoun appears in dual form, it is interpreted as referring to precisely two of the entities (objects or persons) identified by the noun or pronoun. Verbs can also have dual agreement forms in these languages.
- RDF Type:
- owl:Thing
- telix:Number
[back to top]
Instance: telix:future
URI: http://purl.org/telix#future
Future tense - The future tense marks the event described by a given verb as not having happened yet, but expected to happen in the future (in an absolute tense system), or to happen subsequent to some other event, whether that is past, present, or future (in a relative tense system).
- RDF Type:
- owl:Thing
- telix:Tense
[back to top]
Instance: telix:jussive
URI: http://purl.org/telix#jussive
Jussive - The jussive mood, found in Arabic for instance, is used to express pleading, insistence, imploring, self-encouragement, wishing, desiring, intention, commanding, purpose or consequence.
- RDF Type:
- telix:Mood
- owl:Thing
[back to top]
Instance: telix:past
URI: http://purl.org/telix#past
Past tense - The past tense is a grammatical tense that places an action or situation in the past of the current moment or prior to some other event, whether that is past, present, or future.
- RDF Type:
- owl:Thing
- telix:Tense
[back to top]
Instance: telix:subjunctive
URI: http://purl.org/telix#subjunctive
Subjunctive - The subjunctive is a verb mood typically used in subordinate clauses to express various states of irreality such as wish, emotion, possibility, judgment, opinion, necessity, or action that has not yet occurred.
- RDF Type:
- owl:Thing
- telix:Mood
[back to top]
Instance: telix:trial
URI: http://purl.org/telix#trial
Trial number - As in some Austronesian languages, it is a grammatical number referring to 'three items', in contrast to 'singular' (one item), 'dual' (two items), and 'plural' (four or more items).
- RDF Type:
- telix:Number
- owl:Thing
[back to top]
...
Appendix: Reference
The URL of the TELIX ontology is FIXME.
