ISO 24617-1:2012
(Main)Language resource management — Semantic annotation framework (SemAF) — Part 1: Time and events (SemAF-Time, ISO-TimeML)
Language resource management — Semantic annotation framework (SemAF) — Part 1: Time and events (SemAF-Time, ISO-TimeML)
Temporal information in natural language texts is an increasingly important component to the understanding of those texts. ISO 24617-1:2012, SemAF-Time, specifies a formalized XML-based markup language called ISO-TimeML, with a systematic way to extract and represent temporal information, as well as to facilitate the exchange of temporal information, both between operational language processing systems and between different temporal representation schemes. The use of guidelines for temporal annotation has been fully attested with examples from the TimeBank corpus, a collection of 183 documents that have been annotated by TimeML before the current version of ISO-TimeML was formulated.
Gestion des ressources langagières — Cadre d'annotation sémantique (SemAF) — Partie 1: Temps et événements (SemAF-Time, ISO-TimeML)
Upravljanje z jezikovnimi viri - Ogrodje za semantično označevanje (SemAF) - 1. del: Čas in dogodki (SemAF-Time, ISO-TimeML)
Časovne informacije v besedilih v naravnem jeziku so vedno pomembnejše za razumevanje teh besedil. Ta del standarda ISO 24617, SemAF-Time, določa formaliziran jezik za označevanje ISO-TimeML, ki temelji na XML in na sistematičen način izlušči in predstavi časovne informacije ter olajšuje izmenjavo časovnih informacij med sistemi obdelave izvajalnega jezika in med različnimi shemami časovne predstavitve. Uporaba smernic za časovno označevanje je potrjena s primeri iz korpusa TimeBank, zbirko 183 dokumentov, ki so bili označeni s TimeML pred oblikovanjem trenutne različice ISO-TimeML.
General Information
Standards Content (Sample)
SLOVENSKI STANDARD
01-julij-2013
Upravljanje z jezikovnimi viri - Ogrodje za semantično označevanje (SemAF) - 1.
del: Čas in dogodki (SemAF-Time, ISO-TimeML)
Language resource management -- Semantic annotation framework (SemAF) -- Part 1:
Time and events (SemAF-Time, ISO-TimeML)
Gestion des ressources langagières -- Cadre d'annotation sémantique (SemAF) -- Partie
1: Temps et événements (SemAF-Time, ISO-TimeML)
Ta slovenski standard je istoveten z: ISO 24617-1:2012
ICS:
01.020 Terminologija (načela in Terminology (principles and
koordinacija) coordination)
01.140.20 Informacijske vede Information sciences
35.240.30 Uporabniške rešitve IT v IT applications in information,
informatiki, dokumentiranju in documentation and
založništvu publishing
2003-01.Slovenski inštitut za standardizacijo. Razmnoževanje celote ali delov tega standarda ni dovoljeno.
INTERNATIONAL ISO
STANDARD 24617-1
First edition
2012-01-15
Language resource management —
Semantic annotation framework
(SemAF) —
Part 1:
Time and events (SemAF-Time,
ISO-TimeML)
Gestion des ressources langagières — Cadre d'annotation sémantique
(SemAF) —
Partie 1: Temps et événements (SemAF-Time, ISO-TimeML)
Reference number
©
ISO 2012
© ISO 2012
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized in any form or by any means,
electronic or mechanical, including photocopying and microfilm, without permission in writing from either ISO at the address below or
ISO's member body in the country of the requester.
ISO copyright office
Case postale 56 CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Published in Switzerland
ii © ISO 2012 – All rights reserved
Contents Page
Foreword . vi
Introduction . vii
1 Scope . 1
2 Normative references . 1
3 Terms and definitions . 1
4 Overview . 4
5 Motivation and requirements . 4
6 Basic concepts and metamodel . 5
7 Specification of ISO-TimeML . 8
7.1 Overview . 8
7.2 Abstract syntax . 8
7.2.1 Introduction . 8
7.2.2 Conceptual inventory . 9
7.2.3 Syntax rules . 9
7.3 Concrete XML-based syntax . 10
7.3.1 TimeML vs. ISO-TimeML: Stand-off annotation and other differences . 10
7.3.2 Naming conventions . 12
7.3.3 Example annotations . 12
7.3.4 Basic elements: , , and . 12
7.3.5 Link elements: , , and . 18
7.3.6 Other tags: , and . 22
8 Towards a semantics for ISO-TimeML . 26
8.1 Overview . 26
8.2 Tense and aspect in language . 26
8.2.1 Tense . 26
8.2.2 Aspect . 26
8.3 Temporal relations . 27
8.4 An interval-based semantics for ISO-TimeML . 28
8.4.1 Technical preliminaries for interval temporal logic . 28
8.4.2 Basic event-structure . 29
8.4.3 The interpretation of . 31
8.4.4 Interpretive rule summary . 36
8.5 An event-based semantics for ISO-TimeML . 37
8.5.1 Introduction . 37
8.5.2 Defining an event-based semantics . 38
Annex A (normative) Core annotation guidelines . 41
A.1 Introduction . 41
A.2 ISO-TimeML elements and their attributes . 41
A.2.1 The element . 41
A.2.2 The element . 48
A.2.3 The element . 55
A.3 The link elements: , , and . 56
A.3.1 Overview . 56
A.3.2 The element . 56
A.3.3 The element . 59
A.3.4 The element . 61
A.3.5 The element . 62
Annex B (informative) Completely annotated examples .63
B.1 Complex TIMEX3 examples .63
B.2 Complex TLINK and SLINK examples .64
B.3 Causative examples .67
Annex C (informative) Event and temporal annotations for Chinese .68
Annex D (informative) Annotation for Italian fragment .74
D.1 Introduction .74
D.2 Basic references .74
D.3 ISO-TimeML elements and their attributes .74
D.3.1 How to annotate EVENTs .74
D.3.2 Event identification and tag span .75
D.3.3 What NOT to tag .78
D.3.4 Introductory note .78
D.4 The element .81
D.5 The link tags .82
D.6 Informative: Examples of tense, aspect and mood annotation in Italian.82
D.7 Sample of Italian annotation .84
Annex E (informative) Temporal annotation of predicates in Korean .89
E.1 Introduction .89
E.2 Basic references .89
E.3 Morpholoy of Korean predicates .89
E.4 Temporal structure: informative .91
E.5 Temporal annotation of non-Latin texts .92
E.6 Tense .93
E.6.1 Tense markers .93
E.6.2 Annotation guidelines for the attribute @tense .95
E.6.3 Contextual interpretation of tense .96
E.7 Aspect . 114
E.7.1 Aspect markers . 114
E.7.2 Annotation of aspect markers . 115
E.7.3 Interpretation of aspectual features. 116
E.7.4 Interpretation conditions of aspect . 118
E.8 Modality . 118
E.8.1 Conjectural modal markers . 118
E.8.2 Annotation of modality CONJECTURAL . 119
E.8.3 Interpretation of modality CONJECTURAL . 120
E.9 Mood . 120
E.9.1 Mood markers . 120
E.9.2 Annotation of mood RETROSPECTIVE . 121
E.9.3 Interpretation of RETROSPECTIVE mood . 122
E.10 Specific values for attributes in Korean . 122
E.11 Summary . 122
Annex F (informative) Past and current activities on temporal and event annotation . 124
F.1 Introductory remarks . 124
F.2 Annotating temporal expressions . 124
F.3 Annotating events . 125
F.4 Annotating relations between times and events . 127
F.4.1 Ways of capturing time-event relational information . 127
F.4.2 Subordinating and aspectual relations . 129
Annex G (informative) Tools and templates . 130
G.1 Overview . 130
G.2 Annotation tools and templates . 130
G.2.1 Overview . 130
G.2.2 The ALEMBIC workbench . 131
G.2.3 The CALLISTO toolkit . 131
G.2.4 The TANGO temporal relation editor . 131
G.3 Analytic tools . 132
iv © ISO 2012 – All rights reserved
G.3.1 Overview . 132
G.3.2 The TARSQI toolkit . 132
G.3.3 The IBM TimeML annotator . 133
G.3.4 The Amsterdam temporal component extractor . 133
G.3.5 The Time Calculus analyser . 133
Annex H (normative) Specification . 134
H.1 Requirement . 134
H.2 Attribute classes . 134
H.2.1 att.anchored . 134
H.2.2 att.annotate . 135
H.2.3 att.id . 135
H.2.4 att.lang . 135
H.2.5 att.linguistic . 136
H.2.6 att.pointing . 138
H.2.7 att.typed . 138
H.3 Elements . 139
H.3.1 . 139
H.3.2 . 139
H.3.3 . 140
H.3.4 . 141
H.3.5 . 141
H.3.6 . 142
H.3.7 . 143
H.3.8 . 145
H.3.9
H.3.10 . 146
H.3.11
H.3.12 . 146
Bibliography . 147
Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards bodies
(ISO member bodies). The work of preparing International Standards is normally carried out through ISO
technical committees. Each member body interested in a subject for which a technical committee has been
established has the right to be represented on that committee. International organizations, governmental and
non-governmental, in liaison with ISO, also take part in the work. ISO collaborates closely with the
International Electrotechnical Commission (IEC) on all matters of electrotechnical standardization.
International Standards are drafted in accordance with the rules given in the ISO/IEC Directives, Part 2.
The main task of technical committees is to prepare International Standards. Draft International Standards
adopted by the technical committees are circulated to the member bodies for voting. Publication as an
International Standard requires approval by at least 75 % of the member bodies casting a vote.
Attention is drawn to the possibility that some of the elements of this document may be the subject of patent
rights. ISO shall not be held responsible for identifying any or all such patent rights.
ISO 24617-1 was prepared by Technical Committee ISO/TC 37, Terminology and other language and content
resources, Subcommittee SC 4, Language resource management.
ISO 24617 consists of the following parts, under the general title Language resource management —
Semantic annotation framework (SemAF):
Part 1: Time and events (SemAF-Time, ISO-TimeML)
Part 2: Dialogue acts
The following parts are under preparation:
Part 4: Semantic roles (SemAF-SRL)
Part 5: Discourse structure (SemAF-DS)
The following parts are planned:
Part 3: Named entities (SemAF-NE)
Part 6: Principles of semantic annotation
Part 7: Spatial information (ISO-Space)
Part 8: Relations in Discourse (SemAF-DRel)
vi © ISO 2012 – All rights reserved
Introduction
This part of ISO 24617 results from the agreement between the TimeML Working Group and the ISO Working
Group, ISO/TC 37/SC 4/WG 2, Language resource management – Semantic annotation, that a joint activity
should take place to accommodate the two existing documents for annotating temporal information,
TimeML 1.2.1 and TimeML Annotation Guidelines, into ISO international standards. This work should lead to
the achievement of two objectives:
modification of the two documents in conformance to the ISO International Standards;
verification of the annotation guidelines for a wide coverage of multilingual resources.
It should be noted that this part of ISO 24617 provides normative guidelines not just for temporal information,
but also for information content in various types of events in English as well as other languages.
INTERNATIONAL STANDARD ISO 24617-1:2012(E)
Language resource management — Semantic annotation
framework (SemAF) —
Part 1:
Time and events (SemAF-Time, ISO-TimeML)
1 Scope
Temporal information in natural language texts is an increasingly important component to the understanding of
those texts. This part of ISO 24617, SemAF-Time, specifies a formalized XML-based markup language called
ISO-TimeML, with a systematic way to extract and represent temporal information, as well as to facilitate the
exchange of temporal information, both between operational language processing systems and between
different temporal representation schemes. The use of guidelines for temporal annotation has been fully
attested with examples from the TimeBank corpus, a collection of 183 documents that have been annotated
by TimeML before the current version of ISO-TimeML was formulated.
NOTE Throughout this document, SemAF-Time refers to the ISO 24617-1, while ISO-TimeML refers to the
annotation language specified in this document.
2 Normative references
The following referenced documents are indispensable for the application of this document. For dated
references, only the edition cited applies. For undated references, the latest edition of the referenced
document (including any amendments) applies.
NOTE The first reference shows how dates and times are represented and the second provides a format for the
standoff representation of ISO-TimeML annotation presented here.
ISO 8601:2004, Data elements and interchange formats — Information interchange — Representation of
dates and times
ISO 24612:2011, Language resource management — Linguistic annotation framework (LAF)
3 Terms and definitions
For the purposes of this document, the terms and definitions in ISO 8601:2004 and the following apply.
NOTE The terms and definitions provided below are provided to clarify the terminology relating to the metamodel,
specification, and semantics of ISO-TimeML. Terminology derived from XML and other formal languages as well as from
general temporal logics is not defined here.
3.1
ALINK
linking tag that represents a phase relation between an aspectual verb (or morpheme) and a predicate
denoting an event (3.5)
3.2
annotation
process of adding information to segments of language data or that information itself
3.3
beginning
instant (3.6) at which a temporal interval (3.17) begins
NOTE Adapted from Hobbs and Pan (2004).
3.4
end
instant (3.6) at which a temporal interval (3.17) ends
NOTE Adapted from Hobbs and Pan (2004).
3.5
event
eventuality
something that can be said to obtain or hold true, to happen or to occur
NOTE The term “event” is used here with a very broad notion of event, which includes all kinds of actions, states,
processes, etc. It is not to be confused with the more narrow notion of event as something that happens at a certain point
in time (such as the clock striking 2, or waking up) or during a short period of time (such as laughing).
3.6
instant
point in time with no interior points
NOTE Time is often viewed as a straight line from minus infinity to plus infinity. In this view, time is formed by an
infinite sequence of points. An instant can also be seen as an infinitesimally small interval. Cf. OWL-Time Ontology for
“instant”: http://www.w3.org/TR/owl-time/.
3.7
markable
entity in general, or segment of a text in particular, that is subject to an annotation (3.2)
3.8
MLINK
linking tag that represents the measurement of the duration of an event (3.5) or the measurement of the
length of a (possibly discontinuous) time span
3.9
point of event
instant (3.6) at which the event (3.5) mentioned in a given utterance occurs
NOTE Next to a point of speech, a point of event also needs to be defined in order to interpret tense. For example, in
“Arthur smiled”, the temporal location of the point of event can be defined as being prior to the point of speech.
3.10
point of reference
instant (3.6) of temporal perspective on the event (3.5) in a given utterance
NOTE 1 “Arthur will have gone by tomorrow”, where the point of speech is now, the point of event is some time in the
future, but before the point of reference referred to by “tomorrow”.
NOTE 2 To locate certain tenses in time, a third anchor point is also required, defined as the point of reference.
3.11
point of speech
time unit (3.17) at which a given utterance occurs
NOTE 1 The notion of point of speech is needed in order to interpret tense. This requires the use of anchor points in
time, of which the point of speech is one (point of text, see 3.12, is another one). For example, in “Arthur smiled”, the point
of speech is the time that the utterance is made.
NOTE 2 For a document as a whole, this may be considered to be the same as the document creation time.
2 © ISO 2012 – All rights reserved
3.12
point of text
instant (3.6) at which reported speech is anchored
NOTE It is the point of time considered in the text of the speech. So for example, when a person is telling a story, it is
not enough to know the point of the speech itself (the document creation time), but the point at which the speech in the
story is taking place.
3.13
representation
format in which an annotation (3.2) is rendered, for instance in XML, independent of its content
3.14
SLINK
linking tag that represents a subordinating relation between two events (3.5)
3.15
temporal interval
period
uninterrupted stretch of time, with internal point structure.
NOTE 1 Adapted from WordNet.
NOTE 2 Time is often viewed as a straight line from minus infinity to plus infinity. A temporal interval is a part of that
line without any holes, containing all the points between its beginning and its end.
NOTE 3 In mathematics, an important issue is whether an interval includes its beginning and its end (is “closed”) or not
(is “open” or “half-open”). In natural language descriptions of intervals this may also be relevant, as when describing an
interval in terms of a number of days, but not with the same granularity as in mathematics. Cf. OWL-Time Ontology for
“interval”: http://www.w3.org/TR/owl-time/.
3.16
temporal ordering relation
relation that determines how objects are ordered in time
EXAMPLE precedence, simultaneity.
NOTE There is a limited number of ways to order objects which are collectively called ordering relations.
3.17
temporal unit
element in a time amount (3.18) that quantifies the length of a temporal interval (3.15) or a set of temporal
intervals (3.15)
NOTE 1 Adapted from Bunt (1985).
NOTE 2 In measurement systems, various units are defined for different purposes. Small units such as seconds and
minutes are defined to measure small temporal intervals; as one may want to avoid working with big numbers, for larger
temporal intervals, units such as week, year, decade, and century are defined.
NOTE 3 The amount of a temporal unit is called a measure.
3.18
time amount
quantity of time, measured by temporal units (3.17) over temporal intervals (3.15)
NOTE 1 Adapted from Bunt (1985).
NOTE 2 A time amount is a measure of time that can be expressed in terms of a number of temporal units, such as
“half an hour” or “30 minutes”.
3.19
tense
way that languages express the time at which an event (3.5) described by a sentence occurs
NOTE This is characterized as a property of a verb form. Noun forms will not be said to exhibit tense but rather
temporal markers.
3.20
TLINK
linking tag that represents a temporal relation between two temporal entities: namely, between two events
(3.5), two temporal expressions, or between a temporal expression and an event
NOTE 1 Adapted from Pustejovsky et al. (2004).
NOTE 2 Some ordering relations cannot be expressed by an ordering relation between two events because a signal, like
a temporal preposition, complicates the ordering or there is an ordering relation between a temporal signal and an event.
4 Overview
An understanding of temporal information is needed to better understand natural language texts in general.
Previous work in time stamping is a step in the right direction, but to fully appreciate the complexity of a text with
respect to time, the ability to order events and temporal expressions is needed. This part of ISO 24617 defines
ISO-TimeML, a markup language for time and events, which has been specifically designed for this task.
ISO-TimeML annotates all expressions having temporal import, broadly categorized as temporal expressions
and eventualities (situations, events, states, and activities). Temporal expressions and events participate in
temporal relationships (e.g. “before”, “simultaneous”), subordinating relationships (e.g. “intensional”, “factive”),
and aspectual relationships (e.g. “initiates”, “continues”). ISO-TimeML provides an additional expressive
capability of capturing and representing the complexities of these relationships.
TimeML, the precursor of ISO-TimeML, is already in use in a number of applications focusing on analysis
(manual and automatic) of news articles. The TimeBank corpus contains approximately 185 such documents
and has been validated against the most recent version of TimeML. The resulting output of a TimeML
annotated document is in XML, which allows for general XML validation methods to be used. In addition to
supporting interoperability among different temporal representation schemes, TimeML has been shown
adequate to support a mapping from the temporal information in a text to its formal representation in a Web
Ontology Language such as OWL-Time.
Unlike prior event annotation schemes, ISO-TimeML's somewhat unique definition of an event does not limit
the standard's applicability to specific natural language genres. An ISO-TimeML event is simply something
that can be related to another event or temporal expression using an ISO-TimeML relationship — thus an
ISO-TimeML-compliant representation can be adapted (derived) from the full standard specification,
appropriate to different genres, styles, domains, and applications. Future work will involve applying the
standard in such different contexts, and formulating guidelines and principles for appropriate use of
ISO-TimeML in a variety of language engineering environments.
5 Motivation and requirements
The identification of temporal and event expressions in natural language text is a critical component of any
robust information retrieval or language understanding system, and recently this has become an area of
intense research in computational linguistics and Artificial Intelligence. The importance of temporal awareness
to question answering systems has become more obvious as current systems strive to move beyond keyword
and simple named-entity extraction. Named-entity recognition has moved the fields of information retrieval and
information exploitation closer to access by content, by allowing some identification of names, locations and
products in texts. One of the major problems that has not been solved is the recognition of events and their
temporal anchorings in text. Events are naturally anchored in time within a narrative. Without a robust ability to
identify and extract events and their temporal anchoring from a text, the real aboutness of the text can be
missed. Moreover, since entities and their properties change over time, a database of assertions about
entities will be incomplete or incorrect if it does not capture how these properties are temporally updated. To
this end, event recognition drives basic inferences from text.
4 © ISO 2012 – All rights reserved
As it happens, however, much of the temporal information in an article or narrative is left implicit in the text.
The exact temporal designation of events is rarely explicit and many temporal expressions are vague at best.
A crucial first step in the automatic extraction of information from such texts, for use in applications such as
automatic question answering or summarization, is the capacity to identify what events are being described
and to make explicit when these events occurred.
Another important point is that, although most of the information on the web is in natural language, it is
unlikely that it will ever be marked up for semantic retrieval, if that entails hand annotation. Natural language
programs will have to process the contents of web pages to produce annotations. Remarkable progress has
been made in the last decade in the use of statistical techniques for analysing text. However, these
techniques, for the most part, depend on having large amounts of annotated data, and annotations require an
annotation scheme. Hence, in addition to developing the necessary tools for temporal analysis, it is important
to enable for seamless integration into existing and emerging ontologies, such as OWL. Interest in temporal
analysis and event-based reasoning has contributed to the development of a specification language for events
and temporal expressions and their orderings (TimeML). Some issues relating to temporal and event
identification have remained unresolved, however, and ISO-TimeML has been designed to address these
issues. Specifically, four basic problems in event-temporal identification have been addressed in the design of
ISO-TimeML:
time anchoring of events (identifying an event and anchoring it in time);
ordering events with respect to one another (distinguishing lexical from discourse properties of temporal
ordering);
reasoning with contextually underspecified temporal expressions (temporal functions such as “last week”
and “two weeks before”);
reasoning about the persistence of events (how long does an event or the outcome of an event last).
The specification language, ISO-TimeML, is designed to address these issues, in addition to handling basic
tense and aspect features.
Linking a formal theory of time with an annotation scheme aimed at extracting rich temporal information from
natural language text is significant for at least two reasons. It will allow us to use the multitude of temporal
facts expressed in text as the ground propositions in a system for reasoning about temporal relations. It will
also constitute a forcing function for developing the coverage of a temporal reasoning system, as we
encounter phenomena not normally covered by such systems, such as complex descriptions of temporal
aggregates.
6 Basic concepts and metamodel
Regarding the temporal information in a document, a distinction can be made between (1) the temporal
metadata, regarding when the document was created, published, distributed, received, revised, etc., and (2)
the temporal properties of the events and situations that are described in the document. The former type of
information is associated with the document as a whole; information of the latter type will be associated in
annotations with parts of the text in the document, “markables” such as words and phrases.
Temporal objects and relations have been studied from logical and ontological points of view; well-known
studies include those by Allen (1984), Prior (1967), and more recently Hobbs and Pan (2004); see also the
collection of papers in Mani et al. (2005). The most common view of time, which underlies most natural
languages, is that time is an unbounded linear space running from a metaphorical “beginning of time” at minus
infinity to an equally metaphorical “end of time” at plus infinity. This linear space can be represented as a
straight line, the points of which correspond to moments in time; following Hobbs and Pan (2004), we will also
use the term “instant” to refer to time points. From a mathematical point of view, the points on the time line are
line segments of infinitesimally small size, corresponding to the intuition that a moment in time can, in principle,
be determined with any precision that one may wish.
For linguistic and philosophical reasons, several classifications have been proposed of verbs describing
various types of states or events, the Vendler classification being the best known (Vendler, 1967). For the
annotation of temporal information in text, not only verbs with their tenses and temporal modifications should
be considered, but also nouns, since nouns may also denote events and situations (“The meeting tomorrow”;
“The six o'clock news”). In TimeML, Pustejovsky et al. (2007) have proposed a classification of states and
events into seven categories. In the literature, a distinction is often made between events and states, where
events are commonly characterized as occurring at a point in time or during a certain definite interval,
whereas states may obtain for any indefinite stretch of time (“The Mediterranean Sea separates Europe from
Africa”). On a terminological note: the term “event” will henceforth be used as a generic term that also covers
such notions as “state”, “situation”, “action”, “process”, etc.; this broad notion of event has also been termed
“eventuality” (Bach, 1986).
In reality, nothing happens in infinitesimally small time; every event or state that occurs in reality (or in
someone's mind) requires more than zero time, although natural languages offer speakers the possibility to
express themselves as if something occurs at a precise instant (such as “I will call you at twelve o'clock”).
Since instants are formally a special kind of interval, a consistent approach to modelling the time that an event
occurs is to always use intervals, where it may happen that the interval associated with a particular event is
regarded as having zero length, and thus being an instant. This is reflected in the metamodel presented in
Figure 1, which uniformly relates events with temporal intervals.
The length of an interval can also occur as temporal information in a text, as in “I used twelve hours to read
that book” and “It takes seven minutes to walk to the station”. An expression such as “seven minutes” does
not denote an interval, but the length of an interval. It is the temporal equivalent of spatial distance (“seven
miles”). To describe the length of a temporal interval, one needs a unit of measurement, which may be
combined with a numerical expression to obtain an amount of time. The metamodel presented below therefore
includes the concept of an amount of time, related to intervals through the function length, and the auxiliary
concepts of temporal units and real numbers. (Moreover, in the ISO-TimeML semantics, different temporal
units are related through a c
...
SLOVENSKI STANDARD
01-julij-2013
8SUDYOMDQMH]MH]LNRYQLPLYLUL2JURGMH]DVHPDQWLþQRR]QDþHYDQMH 6HP$)
GHO ýDVLQGRJRGNL 6HP$)7LPH,627LPH0/
Language resource management -- Semantic annotation framework (SemAF) -- Part 1:
Time and events (SemAF-Time, ISO-TimeML)
Gestion des ressources langagières -- Cadre d'annotation sémantique (SemAF) -- Partie
1: Temps et événements (SemAF-Time, ISO-TimeML)
Ta slovenski standard je istoveten z: ISO 24617-1:2012
ICS:
01.020 7HUPLQRORJLMD QDþHODLQ Terminology (principles and
NRRUGLQDFLMD coordination)
2003-01.Slovenski inštitut za standardizacijo. Razmnoževanje celote ali delov tega standarda ni dovoljeno.
INTERNATIONAL ISO
STANDARD 24617-1
First edition
2012-01-15
Language resource management —
Semantic annotation framework
(SemAF) —
Part 1:
Time and events (SemAF-Time,
ISO-TimeML)
Gestion des ressources langagières — Cadre d'annotation sémantique
(SemAF) —
Partie 1: Temps et événements (SemAF-Time, ISO-TimeML)
Reference number
©
ISO 2012
© ISO 2012
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized in any form or by any means,
electronic or mechanical, including photocopying and microfilm, without permission in writing from either ISO at the address below or
ISO's member body in the country of the requester.
ISO copyright office
Case postale 56 CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Published in Switzerland
ii © ISO 2012 – All rights reserved
Contents Page
Foreword . vi
Introduction . vii
1 Scope . 1
2 Normative references . 1
3 Terms and definitions . 1
4 Overview . 4
5 Motivation and requirements . 4
6 Basic concepts and metamodel . 5
7 Specification of ISO-TimeML . 8
7.1 Overview . 8
7.2 Abstract syntax . 8
7.2.1 Introduction . 8
7.2.2 Conceptual inventory . 9
7.2.3 Syntax rules . 9
7.3 Concrete XML-based syntax . 10
7.3.1 TimeML vs. ISO-TimeML: Stand-off annotation and other differences . 10
7.3.2 Naming conventions . 12
7.3.3 Example annotations . 12
7.3.4 Basic elements: , , and . 12
7.3.5 Link elements: , , and . 18
7.3.6 Other tags: , and . 22
8 Towards a semantics for ISO-TimeML . 26
8.1 Overview . 26
8.2 Tense and aspect in language . 26
8.2.1 Tense . 26
8.2.2 Aspect . 26
8.3 Temporal relations . 27
8.4 An interval-based semantics for ISO-TimeML . 28
8.4.1 Technical preliminaries for interval temporal logic . 28
8.4.2 Basic event-structure . 29
8.4.3 The interpretation of . 31
8.4.4 Interpretive rule summary . 36
8.5 An event-based semantics for ISO-TimeML . 37
8.5.1 Introduction . 37
8.5.2 Defining an event-based semantics . 38
Annex A (normative) Core annotation guidelines . 41
A.1 Introduction . 41
A.2 ISO-TimeML elements and their attributes . 41
A.2.1 The element . 41
A.2.2 The element . 48
A.2.3 The element . 55
A.3 The link elements: , , and . 56
A.3.1 Overview . 56
A.3.2 The element . 56
A.3.3 The element . 59
A.3.4 The element . 61
A.3.5 The element . 62
Annex B (informative) Completely annotated examples .63
B.1 Complex TIMEX3 examples .63
B.2 Complex TLINK and SLINK examples .64
B.3 Causative examples .67
Annex C (informative) Event and temporal annotations for Chinese .68
Annex D (informative) Annotation for Italian fragment .74
D.1 Introduction .74
D.2 Basic references .74
D.3 ISO-TimeML elements and their attributes .74
D.3.1 How to annotate EVENTs .74
D.3.2 Event identification and tag span .75
D.3.3 What NOT to tag .78
D.3.4 Introductory note .78
D.4 The element .81
D.5 The link tags .82
D.6 Informative: Examples of tense, aspect and mood annotation in Italian.82
D.7 Sample of Italian annotation .84
Annex E (informative) Temporal annotation of predicates in Korean .89
E.1 Introduction .89
E.2 Basic references .89
E.3 Morpholoy of Korean predicates .89
E.4 Temporal structure: informative .91
E.5 Temporal annotation of non-Latin texts .92
E.6 Tense .93
E.6.1 Tense markers .93
E.6.2 Annotation guidelines for the attribute @tense .95
E.6.3 Contextual interpretation of tense .96
E.7 Aspect . 114
E.7.1 Aspect markers . 114
E.7.2 Annotation of aspect markers . 115
E.7.3 Interpretation of aspectual features. 116
E.7.4 Interpretation conditions of aspect . 118
E.8 Modality . 118
E.8.1 Conjectural modal markers . 118
E.8.2 Annotation of modality CONJECTURAL . 119
E.8.3 Interpretation of modality CONJECTURAL . 120
E.9 Mood . 120
E.9.1 Mood markers . 120
E.9.2 Annotation of mood RETROSPECTIVE . 121
E.9.3 Interpretation of RETROSPECTIVE mood . 122
E.10 Specific values for attributes in Korean . 122
E.11 Summary . 122
Annex F (informative) Past and current activities on temporal and event annotation . 124
F.1 Introductory remarks . 124
F.2 Annotating temporal expressions . 124
F.3 Annotating events . 125
F.4 Annotating relations between times and events . 127
F.4.1 Ways of capturing time-event relational information . 127
F.4.2 Subordinating and aspectual relations . 129
Annex G (informative) Tools and templates . 130
G.1 Overview . 130
G.2 Annotation tools and templates . 130
G.2.1 Overview . 130
G.2.2 The ALEMBIC workbench . 131
G.2.3 The CALLISTO toolkit . 131
G.2.4 The TANGO temporal relation editor . 131
G.3 Analytic tools . 132
iv © ISO 2012 – All rights reserved
G.3.1 Overview . 132
G.3.2 The TARSQI toolkit . 132
G.3.3 The IBM TimeML annotator . 133
G.3.4 The Amsterdam temporal component extractor . 133
G.3.5 The Time Calculus analyser . 133
Annex H (normative) Specification . 134
H.1 Requirement . 134
H.2 Attribute classes . 134
H.2.1 att.anchored . 134
H.2.2 att.annotate . 135
H.2.3 att.id . 135
H.2.4 att.lang . 135
H.2.5 att.linguistic . 136
H.2.6 att.pointing . 138
H.2.7 att.typed . 138
H.3 Elements . 139
H.3.1 . 139
H.3.2 . 139
H.3.3 . 140
H.3.4 . 141
H.3.5 . 141
H.3.6 . 142
H.3.7 . 143
H.3.8 . 145
H.3.9
H.3.10 . 146
H.3.11
H.3.12 . 146
Bibliography . 147
Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards bodies
(ISO member bodies). The work of preparing International Standards is normally carried out through ISO
technical committees. Each member body interested in a subject for which a technical committee has been
established has the right to be represented on that committee. International organizations, governmental and
non-governmental, in liaison with ISO, also take part in the work. ISO collaborates closely with the
International Electrotechnical Commission (IEC) on all matters of electrotechnical standardization.
International Standards are drafted in accordance with the rules given in the ISO/IEC Directives, Part 2.
The main task of technical committees is to prepare International Standards. Draft International Standards
adopted by the technical committees are circulated to the member bodies for voting. Publication as an
International Standard requires approval by at least 75 % of the member bodies casting a vote.
Attention is drawn to the possibility that some of the elements of this document may be the subject of patent
rights. ISO shall not be held responsible for identifying any or all such patent rights.
ISO 24617-1 was prepared by Technical Committee ISO/TC 37, Terminology and other language and content
resources, Subcommittee SC 4, Language resource management.
ISO 24617 consists of the following parts, under the general title Language resource management —
Semantic annotation framework (SemAF):
Part 1: Time and events (SemAF-Time, ISO-TimeML)
Part 2: Dialogue acts
The following parts are under preparation:
Part 4: Semantic roles (SemAF-SRL)
Part 5: Discourse structure (SemAF-DS)
The following parts are planned:
Part 3: Named entities (SemAF-NE)
Part 6: Principles of semantic annotation
Part 7: Spatial information (ISO-Space)
Part 8: Relations in Discourse (SemAF-DRel)
vi © ISO 2012 – All rights reserved
Introduction
This part of ISO 24617 results from the agreement between the TimeML Working Group and the ISO Working
Group, ISO/TC 37/SC 4/WG 2, Language resource management – Semantic annotation, that a joint activity
should take place to accommodate the two existing documents for annotating temporal information,
TimeML 1.2.1 and TimeML Annotation Guidelines, into ISO international standards. This work should lead to
the achievement of two objectives:
modification of the two documents in conformance to the ISO International Standards;
verification of the annotation guidelines for a wide coverage of multilingual resources.
It should be noted that this part of ISO 24617 provides normative guidelines not just for temporal information,
but also for information content in various types of events in English as well as other languages.
INTERNATIONAL STANDARD ISO 24617-1:2012(E)
Language resource management — Semantic annotation
framework (SemAF) —
Part 1:
Time and events (SemAF-Time, ISO-TimeML)
1 Scope
Temporal information in natural language texts is an increasingly important component to the understanding of
those texts. This part of ISO 24617, SemAF-Time, specifies a formalized XML-based markup language called
ISO-TimeML, with a systematic way to extract and represent temporal information, as well as to facilitate the
exchange of temporal information, both between operational language processing systems and between
different temporal representation schemes. The use of guidelines for temporal annotation has been fully
attested with examples from the TimeBank corpus, a collection of 183 documents that have been annotated
by TimeML before the current version of ISO-TimeML was formulated.
NOTE Throughout this document, SemAF-Time refers to the ISO 24617-1, while ISO-TimeML refers to the
annotation language specified in this document.
2 Normative references
The following referenced documents are indispensable for the application of this document. For dated
references, only the edition cited applies. For undated references, the latest edition of the referenced
document (including any amendments) applies.
NOTE The first reference shows how dates and times are represented and the second provides a format for the
standoff representation of ISO-TimeML annotation presented here.
ISO 8601:2004, Data elements and interchange formats — Information interchange — Representation of
dates and times
ISO 24612:2011, Language resource management — Linguistic annotation framework (LAF)
3 Terms and definitions
For the purposes of this document, the terms and definitions in ISO 8601:2004 and the following apply.
NOTE The terms and definitions provided below are provided to clarify the terminology relating to the metamodel,
specification, and semantics of ISO-TimeML. Terminology derived from XML and other formal languages as well as from
general temporal logics is not defined here.
3.1
ALINK
linking tag that represents a phase relation between an aspectual verb (or morpheme) and a predicate
denoting an event (3.5)
3.2
annotation
process of adding information to segments of language data or that information itself
3.3
beginning
instant (3.6) at which a temporal interval (3.17) begins
NOTE Adapted from Hobbs and Pan (2004).
3.4
end
instant (3.6) at which a temporal interval (3.17) ends
NOTE Adapted from Hobbs and Pan (2004).
3.5
event
eventuality
something that can be said to obtain or hold true, to happen or to occur
NOTE The term “event” is used here with a very broad notion of event, which includes all kinds of actions, states,
processes, etc. It is not to be confused with the more narrow notion of event as something that happens at a certain point
in time (such as the clock striking 2, or waking up) or during a short period of time (such as laughing).
3.6
instant
point in time with no interior points
NOTE Time is often viewed as a straight line from minus infinity to plus infinity. In this view, time is formed by an
infinite sequence of points. An instant can also be seen as an infinitesimally small interval. Cf. OWL-Time Ontology for
“instant”: http://www.w3.org/TR/owl-time/.
3.7
markable
entity in general, or segment of a text in particular, that is subject to an annotation (3.2)
3.8
MLINK
linking tag that represents the measurement of the duration of an event (3.5) or the measurement of the
length of a (possibly discontinuous) time span
3.9
point of event
instant (3.6) at which the event (3.5) mentioned in a given utterance occurs
NOTE Next to a point of speech, a point of event also needs to be defined in order to interpret tense. For example, in
“Arthur smiled”, the temporal location of the point of event can be defined as being prior to the point of speech.
3.10
point of reference
instant (3.6) of temporal perspective on the event (3.5) in a given utterance
NOTE 1 “Arthur will have gone by tomorrow”, where the point of speech is now, the point of event is some time in the
future, but before the point of reference referred to by “tomorrow”.
NOTE 2 To locate certain tenses in time, a third anchor point is also required, defined as the point of reference.
3.11
point of speech
time unit (3.17) at which a given utterance occurs
NOTE 1 The notion of point of speech is needed in order to interpret tense. This requires the use of anchor points in
time, of which the point of speech is one (point of text, see 3.12, is another one). For example, in “Arthur smiled”, the point
of speech is the time that the utterance is made.
NOTE 2 For a document as a whole, this may be considered to be the same as the document creation time.
2 © ISO 2012 – All rights reserved
3.12
point of text
instant (3.6) at which reported speech is anchored
NOTE It is the point of time considered in the text of the speech. So for example, when a person is telling a story, it is
not enough to know the point of the speech itself (the document creation time), but the point at which the speech in the
story is taking place.
3.13
representation
format in which an annotation (3.2) is rendered, for instance in XML, independent of its content
3.14
SLINK
linking tag that represents a subordinating relation between two events (3.5)
3.15
temporal interval
period
uninterrupted stretch of time, with internal point structure.
NOTE 1 Adapted from WordNet.
NOTE 2 Time is often viewed as a straight line from minus infinity to plus infinity. A temporal interval is a part of that
line without any holes, containing all the points between its beginning and its end.
NOTE 3 In mathematics, an important issue is whether an interval includes its beginning and its end (is “closed”) or not
(is “open” or “half-open”). In natural language descriptions of intervals this may also be relevant, as when describing an
interval in terms of a number of days, but not with the same granularity as in mathematics. Cf. OWL-Time Ontology for
“interval”: http://www.w3.org/TR/owl-time/.
3.16
temporal ordering relation
relation that determines how objects are ordered in time
EXAMPLE precedence, simultaneity.
NOTE There is a limited number of ways to order objects which are collectively called ordering relations.
3.17
temporal unit
element in a time amount (3.18) that quantifies the length of a temporal interval (3.15) or a set of temporal
intervals (3.15)
NOTE 1 Adapted from Bunt (1985).
NOTE 2 In measurement systems, various units are defined for different purposes. Small units such as seconds and
minutes are defined to measure small temporal intervals; as one may want to avoid working with big numbers, for larger
temporal intervals, units such as week, year, decade, and century are defined.
NOTE 3 The amount of a temporal unit is called a measure.
3.18
time amount
quantity of time, measured by temporal units (3.17) over temporal intervals (3.15)
NOTE 1 Adapted from Bunt (1985).
NOTE 2 A time amount is a measure of time that can be expressed in terms of a number of temporal units, such as
“half an hour” or “30 minutes”.
3.19
tense
way that languages express the time at which an event (3.5) described by a sentence occurs
NOTE This is characterized as a property of a verb form. Noun forms will not be said to exhibit tense but rather
temporal markers.
3.20
TLINK
linking tag that represents a temporal relation between two temporal entities: namely, between two events
(3.5), two temporal expressions, or between a temporal expression and an event
NOTE 1 Adapted from Pustejovsky et al. (2004).
NOTE 2 Some ordering relations cannot be expressed by an ordering relation between two events because a signal, like
a temporal preposition, complicates the ordering or there is an ordering relation between a temporal signal and an event.
4 Overview
An understanding of temporal information is needed to better understand natural language texts in general.
Previous work in time stamping is a step in the right direction, but to fully appreciate the complexity of a text with
respect to time, the ability to order events and temporal expressions is needed. This part of ISO 24617 defines
ISO-TimeML, a markup language for time and events, which has been specifically designed for this task.
ISO-TimeML annotates all expressions having temporal import, broadly categorized as temporal expressions
and eventualities (situations, events, states, and activities). Temporal expressions and events participate in
temporal relationships (e.g. “before”, “simultaneous”), subordinating relationships (e.g. “intensional”, “factive”),
and aspectual relationships (e.g. “initiates”, “continues”). ISO-TimeML provides an additional expressive
capability of capturing and representing the complexities of these relationships.
TimeML, the precursor of ISO-TimeML, is already in use in a number of applications focusing on analysis
(manual and automatic) of news articles. The TimeBank corpus contains approximately 185 such documents
and has been validated against the most recent version of TimeML. The resulting output of a TimeML
annotated document is in XML, which allows for general XML validation methods to be used. In addition to
supporting interoperability among different temporal representation schemes, TimeML has been shown
adequate to support a mapping from the temporal information in a text to its formal representation in a Web
Ontology Language such as OWL-Time.
Unlike prior event annotation schemes, ISO-TimeML's somewhat unique definition of an event does not limit
the standard's applicability to specific natural language genres. An ISO-TimeML event is simply something
that can be related to another event or temporal expression using an ISO-TimeML relationship — thus an
ISO-TimeML-compliant representation can be adapted (derived) from the full standard specification,
appropriate to different genres, styles, domains, and applications. Future work will involve applying the
standard in such different contexts, and formulating guidelines and principles for appropriate use of
ISO-TimeML in a variety of language engineering environments.
5 Motivation and requirements
The identification of temporal and event expressions in natural language text is a critical component of any
robust information retrieval or language understanding system, and recently this has become an area of
intense research in computational linguistics and Artificial Intelligence. The importance of temporal awareness
to question answering systems has become more obvious as current systems strive to move beyond keyword
and simple named-entity extraction. Named-entity recognition has moved the fields of information retrieval and
information exploitation closer to access by content, by allowing some identification of names, locations and
products in texts. One of the major problems that has not been solved is the recognition of events and their
temporal anchorings in text. Events are naturally anchored in time within a narrative. Without a robust ability to
identify and extract events and their temporal anchoring from a text, the real aboutness of the text can be
missed. Moreover, since entities and their properties change over time, a database of assertions about
entities will be incomplete or incorrect if it does not capture how these properties are temporally updated. To
this end, event recognition drives basic inferences from text.
4 © ISO 2012 – All rights reserved
As it happens, however, much of the temporal information in an article or narrative is left implicit in the text.
The exact temporal designation of events is rarely explicit and many temporal expressions are vague at best.
A crucial first step in the automatic extraction of information from such texts, for use in applications such as
automatic question answering or summarization, is the capacity to identify what events are being described
and to make explicit when these events occurred.
Another important point is that, although most of the information on the web is in natural language, it is
unlikely that it will ever be marked up for semantic retrieval, if that entails hand annotation. Natural language
programs will have to process the contents of web pages to produce annotations. Remarkable progress has
been made in the last decade in the use of statistical techniques for analysing text. However, these
techniques, for the most part, depend on having large amounts of annotated data, and annotations require an
annotation scheme. Hence, in addition to developing the necessary tools for temporal analysis, it is important
to enable for seamless integration into existing and emerging ontologies, such as OWL. Interest in temporal
analysis and event-based reasoning has contributed to the development of a specification language for events
and temporal expressions and their orderings (TimeML). Some issues relating to temporal and event
identification have remained unresolved, however, and ISO-TimeML has been designed to address these
issues. Specifically, four basic problems in event-temporal identification have been addressed in the design of
ISO-TimeML:
time anchoring of events (identifying an event and anchoring it in time);
ordering events with respect to one another (distinguishing lexical from discourse properties of temporal
ordering);
reasoning with contextually underspecified temporal expressions (temporal functions such as “last week”
and “two weeks before”);
reasoning about the persistence of events (how long does an event or the outcome of an event last).
The specification language, ISO-TimeML, is designed to address these issues, in addition to handling basic
tense and aspect features.
Linking a formal theory of time with an annotation scheme aimed at extracting rich temporal information from
natural language text is significant for at least two reasons. It will allow us to use the multitude of temporal
facts expressed in text as the ground propositions in a system for reasoning about temporal relations. It will
also constitute a forcing function for developing the coverage of a temporal reasoning system, as we
encounter phenomena not normally covered by such systems, such as complex descriptions of temporal
aggregates.
6 Basic concepts and metamodel
Regarding the temporal information in a document, a distinction can be made between (1) the temporal
metadata, regarding when the document was created, published, distributed, received, revised, etc., and (2)
the temporal properties of the events and situations that are described in the document. The former type of
information is associated with the document as a whole; information of the latter type will be associated in
annotations with parts of the text in the document, “markables” such as words and phrases.
Temporal objects and relations have been studied from logical and ontological points of view; well-known
studies include those by Allen (1984), Prior (1967), and more recently Hobbs and Pan (2004); see also the
collection of papers in Mani et al. (2005). The most common view of time, which underlies most natural
languages, is that time is an unbounded linear space running from a metaphorical “beginning of time” at minus
infinity to an equally metaphorical “end of time” at plus infinity. This linear space can be represented as a
straight line, the points of which correspond to moments in time; following Hobbs and Pan (2004), we will also
use the term “instant” to refer to time points. From a mathematical point of view, the points on the time line are
line segments of infinitesimally small size, corresponding to the intuition that a moment in time can, in principle,
be determined with any precision that one may wish.
For linguistic and philosophical reasons, several classifications have been proposed of verbs describing
various types of states or events, the Vendler classification being the best known (Vendler, 1967). For the
annotation of temporal information in text, not only verbs with their tenses and temporal modifications should
be considered, but also nouns, since nouns may also denote events and situations (“The meeting tomorrow”;
“The six o'clock news”). In TimeML, Pustejovsky et al. (2007) have proposed a classification of states and
events into seven categories. In the literature, a distinction is often made between events and states, where
events are commonly characterized as occurring at a point in time or during a certain definite interval,
whereas states may obtain for any indefinite stretch of time (“The Mediterranean Sea separates Europe from
Africa”). On a terminological note: the term “event” will henceforth be used as a generic term that also covers
such notions as “state”, “situation”, “action”, “process”, etc.; this broad notion of event has also been termed
“eventuality” (Bach, 1986).
In reality, nothing happens in infinitesimally small time; every event or state that occurs in reality (or in
someone's mind) requires more than zero time, although natural languages offer speakers the possibility to
express themselves as if something occurs at a precise instant (such as “I will call you at twelve o'clock”).
Since instants are formally a special kind of interval, a consistent approach to modelling the time that an event
occurs is to always use intervals, where it may happen that the interval associated with a particular event is
regarded as having zero length, and thus being an instant. This is reflected in the metamodel presented in
Figure 1, which uniformly relates events with temporal intervals.
The length of an interval can also occur as temporal information in a text, as in “I used twelve hours to read
that book” and “It takes seven minutes to walk to the station”. An expression such as “seven minutes” does
not denote an interval, but the length of an interval. It is the temporal equivalent of spatial distance (“seven
miles”). To describe the length of a temporal interval, one needs a unit of measurement, which may be
combined with a numerical expression to obtain an amount of time. The metamodel presented below therefore
includes the concept of an amount of time, related to intervals through the function length, and the auxiliary
concepts of temporal units and real numbers. (Moreover, in the ISO-TimeML semantics, different temporal
units are related through a conversion function, stipulating such things as 1 hour = 60 minutes;
1 day = 24 hours, etc. An amount of time can be characterized equivalently by as many pairs ...
INTERNATIONAL ISO
STANDARD 24617-1
First edition
2012-01-15
Language resource management —
Semantic annotation framework
(SemAF) —
Part 1:
Time and events (SemAF-Time,
ISO-TimeML)
Gestion des ressources langagières — Cadre d'annotation sémantique
(SemAF) —
Partie 1: Temps et événements (SemAF-Time, ISO-TimeML)
Reference number
©
ISO 2012
© ISO 2012
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized in any form or by any means,
electronic or mechanical, including photocopying and microfilm, without permission in writing from either ISO at the address below or
ISO's member body in the country of the requester.
ISO copyright office
Case postale 56 CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Published in Switzerland
ii © ISO 2012 – All rights reserved
Contents Page
Foreword . vi
Introduction . vii
1 Scope . 1
2 Normative references . 1
3 Terms and definitions . 1
4 Overview . 4
5 Motivation and requirements . 4
6 Basic concepts and metamodel . 5
7 Specification of ISO-TimeML . 8
7.1 Overview . 8
7.2 Abstract syntax . 8
7.2.1 Introduction . 8
7.2.2 Conceptual inventory . 9
7.2.3 Syntax rules . 9
7.3 Concrete XML-based syntax . 10
7.3.1 TimeML vs. ISO-TimeML: Stand-off annotation and other differences . 10
7.3.2 Naming conventions . 12
7.3.3 Example annotations . 12
7.3.4 Basic elements: , , and . 12
7.3.5 Link elements: , , and . 18
7.3.6 Other tags: , and . 22
8 Towards a semantics for ISO-TimeML . 26
8.1 Overview . 26
8.2 Tense and aspect in language . 26
8.2.1 Tense . 26
8.2.2 Aspect . 26
8.3 Temporal relations . 27
8.4 An interval-based semantics for ISO-TimeML . 28
8.4.1 Technical preliminaries for interval temporal logic . 28
8.4.2 Basic event-structure . 29
8.4.3 The interpretation of . 31
8.4.4 Interpretive rule summary . 36
8.5 An event-based semantics for ISO-TimeML . 37
8.5.1 Introduction . 37
8.5.2 Defining an event-based semantics . 38
Annex A (normative) Core annotation guidelines . 41
A.1 Introduction . 41
A.2 ISO-TimeML elements and their attributes . 41
A.2.1 The element . 41
A.2.2 The element . 48
A.2.3 The element . 55
A.3 The link elements: , , and . 56
A.3.1 Overview . 56
A.3.2 The element . 56
A.3.3 The element . 59
A.3.4 The element . 61
A.3.5 The element . 62
Annex B (informative) Completely annotated examples .63
B.1 Complex TIMEX3 examples .63
B.2 Complex TLINK and SLINK examples .64
B.3 Causative examples .67
Annex C (informative) Event and temporal annotations for Chinese .68
Annex D (informative) Annotation for Italian fragment .74
D.1 Introduction .74
D.2 Basic references .74
D.3 ISO-TimeML elements and their attributes .74
D.3.1 How to annotate EVENTs .74
D.3.2 Event identification and tag span .75
D.3.3 What NOT to tag .78
D.3.4 Introductory note .78
D.4 The element .81
D.5 The link tags .82
D.6 Informative: Examples of tense, aspect and mood annotation in Italian.82
D.7 Sample of Italian annotation .84
Annex E (informative) Temporal annotation of predicates in Korean .89
E.1 Introduction .89
E.2 Basic references .89
E.3 Morpholoy of Korean predicates .89
E.4 Temporal structure: informative .91
E.5 Temporal annotation of non-Latin texts .92
E.6 Tense .93
E.6.1 Tense markers .93
E.6.2 Annotation guidelines for the attribute @tense .95
E.6.3 Contextual interpretation of tense .96
E.7 Aspect . 114
E.7.1 Aspect markers . 114
E.7.2 Annotation of aspect markers . 115
E.7.3 Interpretation of aspectual features. 116
E.7.4 Interpretation conditions of aspect . 118
E.8 Modality . 118
E.8.1 Conjectural modal markers . 118
E.8.2 Annotation of modality CONJECTURAL . 119
E.8.3 Interpretation of modality CONJECTURAL . 120
E.9 Mood . 120
E.9.1 Mood markers . 120
E.9.2 Annotation of mood RETROSPECTIVE . 121
E.9.3 Interpretation of RETROSPECTIVE mood . 122
E.10 Specific values for attributes in Korean . 122
E.11 Summary . 122
Annex F (informative) Past and current activities on temporal and event annotation . 124
F.1 Introductory remarks . 124
F.2 Annotating temporal expressions . 124
F.3 Annotating events . 125
F.4 Annotating relations between times and events . 127
F.4.1 Ways of capturing time-event relational information . 127
F.4.2 Subordinating and aspectual relations . 129
Annex G (informative) Tools and templates . 130
G.1 Overview . 130
G.2 Annotation tools and templates . 130
G.2.1 Overview . 130
G.2.2 The ALEMBIC workbench . 131
G.2.3 The CALLISTO toolkit . 131
G.2.4 The TANGO temporal relation editor . 131
G.3 Analytic tools . 132
iv © ISO 2012 – All rights reserved
G.3.1 Overview . 132
G.3.2 The TARSQI toolkit . 132
G.3.3 The IBM TimeML annotator . 133
G.3.4 The Amsterdam temporal component extractor . 133
G.3.5 The Time Calculus analyser . 133
Annex H (normative) Specification . 134
H.1 Requirement . 134
H.2 Attribute classes . 134
H.2.1 att.anchored . 134
H.2.2 att.annotate . 135
H.2.3 att.id . 135
H.2.4 att.lang . 135
H.2.5 att.linguistic . 136
H.2.6 att.pointing . 138
H.2.7 att.typed . 138
H.3 Elements . 139
H.3.1 . 139
H.3.2 . 139
H.3.3 . 140
H.3.4 . 141
H.3.5 . 141
H.3.6 . 142
H.3.7 . 143
H.3.8 . 145
H.3.9
H.3.10 . 146
H.3.11
H.3.12 . 146
Bibliography . 147
Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards bodies
(ISO member bodies). The work of preparing International Standards is normally carried out through ISO
technical committees. Each member body interested in a subject for which a technical committee has been
established has the right to be represented on that committee. International organizations, governmental and
non-governmental, in liaison with ISO, also take part in the work. ISO collaborates closely with the
International Electrotechnical Commission (IEC) on all matters of electrotechnical standardization.
International Standards are drafted in accordance with the rules given in the ISO/IEC Directives, Part 2.
The main task of technical committees is to prepare International Standards. Draft International Standards
adopted by the technical committees are circulated to the member bodies for voting. Publication as an
International Standard requires approval by at least 75 % of the member bodies casting a vote.
Attention is drawn to the possibility that some of the elements of this document may be the subject of patent
rights. ISO shall not be held responsible for identifying any or all such patent rights.
ISO 24617-1 was prepared by Technical Committee ISO/TC 37, Terminology and other language and content
resources, Subcommittee SC 4, Language resource management.
ISO 24617 consists of the following parts, under the general title Language resource management —
Semantic annotation framework (SemAF):
Part 1: Time and events (SemAF-Time, ISO-TimeML)
Part 2: Dialogue acts
The following parts are under preparation:
Part 4: Semantic roles (SemAF-SRL)
Part 5: Discourse structure (SemAF-DS)
The following parts are planned:
Part 3: Named entities (SemAF-NE)
Part 6: Principles of semantic annotation
Part 7: Spatial information (ISO-Space)
Part 8: Relations in Discourse (SemAF-DRel)
vi © ISO 2012 – All rights reserved
Introduction
This part of ISO 24617 results from the agreement between the TimeML Working Group and the ISO Working
Group, ISO/TC 37/SC 4/WG 2, Language resource management – Semantic annotation, that a joint activity
should take place to accommodate the two existing documents for annotating temporal information,
TimeML 1.2.1 and TimeML Annotation Guidelines, into ISO international standards. This work should lead to
the achievement of two objectives:
modification of the two documents in conformance to the ISO International Standards;
verification of the annotation guidelines for a wide coverage of multilingual resources.
It should be noted that this part of ISO 24617 provides normative guidelines not just for temporal information,
but also for information content in various types of events in English as well as other languages.
INTERNATIONAL STANDARD ISO 24617-1:2012(E)
Language resource management — Semantic annotation
framework (SemAF) —
Part 1:
Time and events (SemAF-Time, ISO-TimeML)
1 Scope
Temporal information in natural language texts is an increasingly important component to the understanding of
those texts. This part of ISO 24617, SemAF-Time, specifies a formalized XML-based markup language called
ISO-TimeML, with a systematic way to extract and represent temporal information, as well as to facilitate the
exchange of temporal information, both between operational language processing systems and between
different temporal representation schemes. The use of guidelines for temporal annotation has been fully
attested with examples from the TimeBank corpus, a collection of 183 documents that have been annotated
by TimeML before the current version of ISO-TimeML was formulated.
NOTE Throughout this document, SemAF-Time refers to the ISO 24617-1, while ISO-TimeML refers to the
annotation language specified in this document.
2 Normative references
The following referenced documents are indispensable for the application of this document. For dated
references, only the edition cited applies. For undated references, the latest edition of the referenced
document (including any amendments) applies.
NOTE The first reference shows how dates and times are represented and the second provides a format for the
standoff representation of ISO-TimeML annotation presented here.
ISO 8601:2004, Data elements and interchange formats — Information interchange — Representation of
dates and times
ISO 24612:2011, Language resource management — Linguistic annotation framework (LAF)
3 Terms and definitions
For the purposes of this document, the terms and definitions in ISO 8601:2004 and the following apply.
NOTE The terms and definitions provided below are provided to clarify the terminology relating to the metamodel,
specification, and semantics of ISO-TimeML. Terminology derived from XML and other formal languages as well as from
general temporal logics is not defined here.
3.1
ALINK
linking tag that represents a phase relation between an aspectual verb (or morpheme) and a predicate
denoting an event (3.5)
3.2
annotation
process of adding information to segments of language data or that information itself
3.3
beginning
instant (3.6) at which a temporal interval (3.17) begins
NOTE Adapted from Hobbs and Pan (2004).
3.4
end
instant (3.6) at which a temporal interval (3.17) ends
NOTE Adapted from Hobbs and Pan (2004).
3.5
event
eventuality
something that can be said to obtain or hold true, to happen or to occur
NOTE The term “event” is used here with a very broad notion of event, which includes all kinds of actions, states,
processes, etc. It is not to be confused with the more narrow notion of event as something that happens at a certain point
in time (such as the clock striking 2, or waking up) or during a short period of time (such as laughing).
3.6
instant
point in time with no interior points
NOTE Time is often viewed as a straight line from minus infinity to plus infinity. In this view, time is formed by an
infinite sequence of points. An instant can also be seen as an infinitesimally small interval. Cf. OWL-Time Ontology for
“instant”: http://www.w3.org/TR/owl-time/.
3.7
markable
entity in general, or segment of a text in particular, that is subject to an annotation (3.2)
3.8
MLINK
linking tag that represents the measurement of the duration of an event (3.5) or the measurement of the
length of a (possibly discontinuous) time span
3.9
point of event
instant (3.6) at which the event (3.5) mentioned in a given utterance occurs
NOTE Next to a point of speech, a point of event also needs to be defined in order to interpret tense. For example, in
“Arthur smiled”, the temporal location of the point of event can be defined as being prior to the point of speech.
3.10
point of reference
instant (3.6) of temporal perspective on the event (3.5) in a given utterance
NOTE 1 “Arthur will have gone by tomorrow”, where the point of speech is now, the point of event is some time in the
future, but before the point of reference referred to by “tomorrow”.
NOTE 2 To locate certain tenses in time, a third anchor point is also required, defined as the point of reference.
3.11
point of speech
time unit (3.17) at which a given utterance occurs
NOTE 1 The notion of point of speech is needed in order to interpret tense. This requires the use of anchor points in
time, of which the point of speech is one (point of text, see 3.12, is another one). For example, in “Arthur smiled”, the point
of speech is the time that the utterance is made.
NOTE 2 For a document as a whole, this may be considered to be the same as the document creation time.
2 © ISO 2012 – All rights reserved
3.12
point of text
instant (3.6) at which reported speech is anchored
NOTE It is the point of time considered in the text of the speech. So for example, when a person is telling a story, it is
not enough to know the point of the speech itself (the document creation time), but the point at which the speech in the
story is taking place.
3.13
representation
format in which an annotation (3.2) is rendered, for instance in XML, independent of its content
3.14
SLINK
linking tag that represents a subordinating relation between two events (3.5)
3.15
temporal interval
period
uninterrupted stretch of time, with internal point structure.
NOTE 1 Adapted from WordNet.
NOTE 2 Time is often viewed as a straight line from minus infinity to plus infinity. A temporal interval is a part of that
line without any holes, containing all the points between its beginning and its end.
NOTE 3 In mathematics, an important issue is whether an interval includes its beginning and its end (is “closed”) or not
(is “open” or “half-open”). In natural language descriptions of intervals this may also be relevant, as when describing an
interval in terms of a number of days, but not with the same granularity as in mathematics. Cf. OWL-Time Ontology for
“interval”: http://www.w3.org/TR/owl-time/.
3.16
temporal ordering relation
relation that determines how objects are ordered in time
EXAMPLE precedence, simultaneity.
NOTE There is a limited number of ways to order objects which are collectively called ordering relations.
3.17
temporal unit
element in a time amount (3.18) that quantifies the length of a temporal interval (3.15) or a set of temporal
intervals (3.15)
NOTE 1 Adapted from Bunt (1985).
NOTE 2 In measurement systems, various units are defined for different purposes. Small units such as seconds and
minutes are defined to measure small temporal intervals; as one may want to avoid working with big numbers, for larger
temporal intervals, units such as week, year, decade, and century are defined.
NOTE 3 The amount of a temporal unit is called a measure.
3.18
time amount
quantity of time, measured by temporal units (3.17) over temporal intervals (3.15)
NOTE 1 Adapted from Bunt (1985).
NOTE 2 A time amount is a measure of time that can be expressed in terms of a number of temporal units, such as
“half an hour” or “30 minutes”.
3.19
tense
way that languages express the time at which an event (3.5) described by a sentence occurs
NOTE This is characterized as a property of a verb form. Noun forms will not be said to exhibit tense but rather
temporal markers.
3.20
TLINK
linking tag that represents a temporal relation between two temporal entities: namely, between two events
(3.5), two temporal expressions, or between a temporal expression and an event
NOTE 1 Adapted from Pustejovsky et al. (2004).
NOTE 2 Some ordering relations cannot be expressed by an ordering relation between two events because a signal, like
a temporal preposition, complicates the ordering or there is an ordering relation between a temporal signal and an event.
4 Overview
An understanding of temporal information is needed to better understand natural language texts in general.
Previous work in time stamping is a step in the right direction, but to fully appreciate the complexity of a text with
respect to time, the ability to order events and temporal expressions is needed. This part of ISO 24617 defines
ISO-TimeML, a markup language for time and events, which has been specifically designed for this task.
ISO-TimeML annotates all expressions having temporal import, broadly categorized as temporal expressions
and eventualities (situations, events, states, and activities). Temporal expressions and events participate in
temporal relationships (e.g. “before”, “simultaneous”), subordinating relationships (e.g. “intensional”, “factive”),
and aspectual relationships (e.g. “initiates”, “continues”). ISO-TimeML provides an additional expressive
capability of capturing and representing the complexities of these relationships.
TimeML, the precursor of ISO-TimeML, is already in use in a number of applications focusing on analysis
(manual and automatic) of news articles. The TimeBank corpus contains approximately 185 such documents
and has been validated against the most recent version of TimeML. The resulting output of a TimeML
annotated document is in XML, which allows for general XML validation methods to be used. In addition to
supporting interoperability among different temporal representation schemes, TimeML has been shown
adequate to support a mapping from the temporal information in a text to its formal representation in a Web
Ontology Language such as OWL-Time.
Unlike prior event annotation schemes, ISO-TimeML's somewhat unique definition of an event does not limit
the standard's applicability to specific natural language genres. An ISO-TimeML event is simply something
that can be related to another event or temporal expression using an ISO-TimeML relationship — thus an
ISO-TimeML-compliant representation can be adapted (derived) from the full standard specification,
appropriate to different genres, styles, domains, and applications. Future work will involve applying the
standard in such different contexts, and formulating guidelines and principles for appropriate use of
ISO-TimeML in a variety of language engineering environments.
5 Motivation and requirements
The identification of temporal and event expressions in natural language text is a critical component of any
robust information retrieval or language understanding system, and recently this has become an area of
intense research in computational linguistics and Artificial Intelligence. The importance of temporal awareness
to question answering systems has become more obvious as current systems strive to move beyond keyword
and simple named-entity extraction. Named-entity recognition has moved the fields of information retrieval and
information exploitation closer to access by content, by allowing some identification of names, locations and
products in texts. One of the major problems that has not been solved is the recognition of events and their
temporal anchorings in text. Events are naturally anchored in time within a narrative. Without a robust ability to
identify and extract events and their temporal anchoring from a text, the real aboutness of the text can be
missed. Moreover, since entities and their properties change over time, a database of assertions about
entities will be incomplete or incorrect if it does not capture how these properties are temporally updated. To
this end, event recognition drives basic inferences from text.
4 © ISO 2012 – All rights reserved
As it happens, however, much of the temporal information in an article or narrative is left implicit in the text.
The exact temporal designation of events is rarely explicit and many temporal expressions are vague at best.
A crucial first step in the automatic extraction of information from such texts, for use in applications such as
automatic question answering or summarization, is the capacity to identify what events are being described
and to make explicit when these events occurred.
Another important point is that, although most of the information on the web is in natural language, it is
unlikely that it will ever be marked up for semantic retrieval, if that entails hand annotation. Natural language
programs will have to process the contents of web pages to produce annotations. Remarkable progress has
been made in the last decade in the use of statistical techniques for analysing text. However, these
techniques, for the most part, depend on having large amounts of annotated data, and annotations require an
annotation scheme. Hence, in addition to developing the necessary tools for temporal analysis, it is important
to enable for seamless integration into existing and emerging ontologies, such as OWL. Interest in temporal
analysis and event-based reasoning has contributed to the development of a specification language for events
and temporal expressions and their orderings (TimeML). Some issues relating to temporal and event
identification have remained unresolved, however, and ISO-TimeML has been designed to address these
issues. Specifically, four basic problems in event-temporal identification have been addressed in the design of
ISO-TimeML:
time anchoring of events (identifying an event and anchoring it in time);
ordering events with respect to one another (distinguishing lexical from discourse properties of temporal
ordering);
reasoning with contextually underspecified temporal expressions (temporal functions such as “last week”
and “two weeks before”);
reasoning about the persistence of events (how long does an event or the outcome of an event last).
The specification language, ISO-TimeML, is designed to address these issues, in addition to handling basic
tense and aspect features.
Linking a formal theory of time with an annotation scheme aimed at extracting rich temporal information from
natural language text is significant for at least two reasons. It will allow us to use the multitude of temporal
facts expressed in text as the ground propositions in a system for reasoning about temporal relations. It will
also constitute a forcing function for developing the coverage of a temporal reasoning system, as we
encounter phenomena not normally covered by such systems, such as complex descriptions of temporal
aggregates.
6 Basic concepts and metamodel
Regarding the temporal information in a document, a distinction can be made between (1) the temporal
metadata, regarding when the document was created, published, distributed, received, revised, etc., and (2)
the temporal properties of the events and situations that are described in the document. The former type of
information is associated with the document as a whole; information of the latter type will be associated in
annotations with parts of the text in the document, “markables” such as words and phrases.
Temporal objects and relations have been studied from logical and ontological points of view; well-known
studies include those by Allen (1984), Prior (1967), and more recently Hobbs and Pan (2004); see also the
collection of papers in Mani et al. (2005). The most common view of time, which underlies most natural
languages, is that time is an unbounded linear space running from a metaphorical “beginning of time” at minus
infinity to an equally metaphorical “end of time” at plus infinity. This linear space can be represented as a
straight line, the points of which correspond to moments in time; following Hobbs and Pan (2004), we will also
use the term “instant” to refer to time points. From a mathematical point of view, the points on the time line are
line segments of infinitesimally small size, corresponding to the intuition that a moment in time can, in principle,
be determined with any precision that one may wish.
For linguistic and philosophical reasons, several classifications have been proposed of verbs describing
various types of states or events, the Vendler classification being the best known (Vendler, 1967). For the
annotation of temporal information in text, not only verbs with their tenses and temporal modifications should
be considered, but also nouns, since nouns may also denote events and situations (“The meeting tomorrow”;
“The six o'clock news”). In TimeML, Pustejovsky et al. (2007) have proposed a classification of states and
events into seven categories. In the literature, a distinction is often made between events and states, where
events are commonly characterized as occurring at a point in time or during a certain definite interval,
whereas states may obtain for any indefinite stretch of time (“The Mediterranean Sea separates Europe from
Africa”). On a terminological note: the term “event” will henceforth be used as a generic term that also covers
such notions as “state”, “situation”, “action”, “process”, etc.; this broad notion of event has also been termed
“eventuality” (Bach, 1986).
In reality, nothing happens in infinitesimally small time; every event or state that occurs in reality (or in
someone's mind) requires more than zero time, although natural languages offer speakers the possibility to
express themselves as if something occurs at a precise instant (such as “I will call you at twelve o'clock”).
Since instants are formally a special kind of interval, a consistent approach to modelling the time that an event
occurs is to always use intervals, where it may happen that the interval associated with a particular event is
regarded as having zero length, and thus being an instant. This is reflected in the metamodel presented in
Figure 1, which uniformly relates events with temporal intervals.
The length of an interval can also occur as temporal information in a text, as in “I used twelve hours to read
that book” and “It takes seven minutes to walk to the station”. An expression such as “seven minutes” does
not denote an interval, but the length of an interval. It is the temporal equivalent of spatial distance (“seven
miles”). To describe the length of a temporal interval, one needs a unit of measurement, which may be
combined with a numerical expression to obtain an amount of time. The metamodel presented below therefore
includes the concept of an amount of time, related to intervals through the function length, and the auxiliary
concepts of temporal units and real numbers. (Moreover, in the ISO-TimeML semantics, different temporal
units are related through a conversion function, stipulating such things as 1 hour = 60 minutes;
1 day = 24 hours, etc. An amount of time can be characterized equivalently by as many pairs temporal unit> as there are temporal units, the equivalence being defined through the numerical conversions
between units [see Bunt (1985)].
Regarding the temporal anchoring of events in time, it may be noted that the association of a temporal interval
with an event does not necessarily mean that the event took place during every moment within that interval.
When someone says “I've been working on my presentation from 8.30 to 12 o'clock”, that presumably does
not mean that the speaker has been working on his presentation for every single moment between 8.30 and
12 o'clock; there must have been interruptions for having some coffee, going to the bathroom, etc. In such a
case it is more accurate to anchor the event at the time span starting at 8.30 and ending at 12 o'clock, a “time
span” being understood as a period of time that may have “holes”, where the event was interrupted. The
metamodel shown in Figure 1 does not distinguish time spans, but reflects the assumption that whether an
event occurs during an interval, with or without any interruptions, can only be decided on a case by case basis,
and is best modelled as a property of the temporal anchoring relation applied to a specific event.
ISO 24612:2011 insists on the use o
...
Questions, Comments and Discussion
Ask us and Technical Secretary will try to provide an answer. You can facilitate discussion about the standard in here.
Loading comments...