ISO 24617-2:2012

SLOVENSKI STANDARD
01-julij-2013
Upravljanje z jezikovnimi viri - Ogrodje za semantično označevanje (SemAF) - 2.
del: Dialogi
Language resource management -- Semantic annotation framework (SemAF) -- Part 2:
Dialogue acts
Gestion des ressources langagières -- Cadre d'annotation sémantique (SemAF) -- Partie
2: Actes de dialogue
Ta slovenski standard je istoveten z: ISO 24617-2:2012
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 24617-2
First edition
2012-09-01
Language resource management —
Semantic annotation framework
(SemAF) —
Part 2:
Dialogue acts
Gestion des ressources langagières — Cadre d'annotation sémantique
(SemAF) —
Partie 2: Actes de dialogue
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 . iv
1  Scope . 1
2  Normative references . 1
3  Terms and definitions . 1
4  Purpose and justification . 5
5  Basic concepts and metamodel . 6
6  Definition of communicative functions . 8
7  Annotation schemes . 9
7.1  Structure of annotation schemes . 9
7.2  Multidimensionality and multifunctionality . 10
7.3  Multidimensionality, clustering and dimensions . 11
7.4  Dimension- specific and general-purpose functions . 11
8  Dialogue segmentation . 13
9  Dimensions . 14
9.1  Task. 15
9.2  Auto-Feedback . 15
9.3  Allo-Feedback . 15
9.4  Turn Management . 15
9.5  Time Management . 16
9.6  Discourse Structuring . 16
9.7  Social Obligations Management . 16
9.8  Own Communication Management . 16
9.9  Partner Communication Management . 16
10  Core dialogue acts . 17
10.1  General-purpose functions . 19
10.2  Dimension-specific functions . 20
10.3  Function qualifiers . 22
11  Dialogue act markup language (DiAML) . 23
11.1  Abstract syntax . 23
11.2  Concrete syntax . 24
12  Principles for extending and restricting the standard . 25
12.1  Main design principles . 25
12.2  Schema extension . 27
12.3  Scheme restriction . 27
Annex A (informative) Annotation guidelines . 29
Annex B (informative) Annotated dialogue examples . 43
Annex C (normative) Formal definition of DiAML . 56
Annex D (normative) DiAML technical schema . 63
Annex E (normative)  Data categories for core concepts . 68
Annex F (informative) Examples of possible additional data categories . 88
Annex G (informative) Concepts in existing schemes . 90
Bibliography . 100
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 24617-2 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:
 Part 1: Time and events (SemAF-Time, ISO-TimeML)
 Part 2: Dialogue acts
The following parts are under preparation:
 Part 3: Named entities (SemAF-NE)
 Part 4: Semantic roles (SemAF-SRL)
 Part 5: Discourse structure (SemAF-DS)
 Part 6: Principles of semantic annotation (SemAF-Basics)
 Part 7: Spatial information (ISO-Space)
 Part 8: Semantic relations in discourse (SemAF-DRel)

iv © ISO 2012 – All rights reserved

INTERNATIONAL STANDARD ISO 24617-2:2012(E)

Language resource management — Semantic annotation
framework (SemAF) —
Part 2:
Dialogue acts
1 Scope
This part of ISO 24617 provides a set of empirically and theoretically well-motivated concepts for dialogue
annotation, a formal language for expressing dialogue annotations — the dialogue act markup language
(DiAML) — and a method for segmenting a dialogue into semantic units. This allows the manual or automatic
annotation of dialogue segments with information about the communicative actions which the participants
perform by their contributions to the dialogue. It supports multidimensional annotation, in which units in
dialogue are viewed as having multiple communicative functions. The DiAML language has an XML-based
representation format and a formal semantics which makes it possible to apply inference to DiAML
representations.
This part of ISO 24617 specifies data categories for reference sets of communicative functions and
dimensions of dialogue analysis and provides principles and guidelines for extending these sets or selecting
coherent subsets of them. Additionally, it provides guidelines for annotators and annotated examples. It is
applicable to spoken, written and multimodal dialogues involving two or more participants.
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.
ISO 12620:2009, Terminology and other language resources — Specification of data categories and
management of a Data Category Registry for language resources
ISO 24610-1:2006, Language resource management — Feature structures — Part 1: Feature structure
representation
ISO 24612:2011, Language resource management — Linguistic annotation framework
3 Terms and definitions
1)
For the purposes of this document, the following terms and definitions apply.

1) In this document, “he”, “him” and “his” are used in a generic sense, without implying any gender-related distinctions.
3.1
addressee
dialogue (3.5) participant (3.13) oriented to by the sender (3.18) in a manner to suggest that his utterances
(3.22) are particularly intended for this participant and that some response is therefore anticipated from this
participant, more so than from the other participants
Note to entry: This definition is a de facto standard in the linguistics literature. It has been slightly modified here, in
replacing “speaker” by “sender” and avoiding the use of ambiguous pronouns. Goffman's original definition says: “dialogue
participant oriented to by the speaker in a manner to suggest that his utterances are particularly intended for him and that
some response is therefore anticipated from him/her, more so than from the other participants”.
[SOURCE: Goffman (1981).]
3.2
allo-feedback act
feedback act (3.8) where the sender (3.18) elicits information about the addressee's (3.1) processing of an
utterance (3.22) that the sender contributed to the dialogue (3.5) or where the sender provides information
about his perceived processing by the addressee of an utterance that the sender contributed to the dialogue
before
EXAMPLE A: Now move up.
B: Slightly northeast you mean?
A: Slightly yeah.
A performs an allo-feedback act signalling that he thinks B understood his first utterance correctly.
3.3
auto-feedback act
feedback act (3.8) where the sender (3.18) provides information about his own processing of an utterance
(3.22) contributed to the dialogue (3.5) by another participant (3.13)
EXAMPLE B's utterance in the example dialogue fragment in (3.2) signals that he is uncertain whether he
understood the previous utterance correctly.
3.4
communicative function
property of certain stretches of communicative behaviour, describing how the behaviour changes the
information state (3.12) of an understander of the behaviour
Note to entry: A communicative function may be “qualified”, i.e. one or more qualifiers (3.14) may be associated with it.
For example, an answer may be qualified as “uncertain” and the acceptance of a request may be “conditional”. See 10.3
for explanation and examples.
3.5
dialogue
exchange of utterances (3.22) between two or more persons or artificial conversational systems
3.6
dialogue act
communicative activity of a dialogue (3.5) participant (3.13), interpreted as having a certain communicative
function (3.4) and semantic content (3.16)
Note to entry: A dialogue act may also have certain functional dependence relations (3.10), rhetorical relations (3.15) and
feedback dependence relations (3.9) with other units in a dialogue (3.5).
3.7
dimension
class of dialogue acts (3.6) that are concerned with a particular aspect of communication, corresponding to a
particular category of semantic content
2 © ISO 2012 – All rights reserved

EXAMPLE Dialogue acts advancing the task or activity that motivates the dialogue (the Task dimension), dialogue
acts providing and eliciting feedback (the Auto- and Allo-Feedback dimensions) and dialogue acts for allocating the
speaker role (the Turn Management dimension).
Note to entry: See Clauses 5, 7 and 9 for discussion and more examples.
3.8
feedback act
dialogue act (3.6) which provides or elicits information about the sender's (3.18) or the addressee's (3.1)
processing of something that was uttered in the dialogue
Note to entry: Two classes of feedback are distinguished in this part of ISO 24617: allo-feedback acts (3.2) and auto-
feedback acts (3.3).
3.9
feedback dependence relation
relation between a feedback act (3.8) and the stretch of communicative behaviour whose processing the act
provides or elicits information about
EXAMPLE In the example that accompanies definition 3.2, both the allo-feedback act expressed by utterance 3 and
the auto-feedback act expressed by utterance 2 have a feedback dependence relation to utterance 1.
3.10
functional dependence relation
relation between a given dialogue act (3.6) and a preceding dialogue act on which the semantic content of the
given dialogue act depends due to its communicative function (3.4)
EXAMPLE The relation between an answer and the corresponding question, such as between utterance 3 and
utterance 2 in the example accompanying definition 3.2; or the relation between the acceptance of an offer and the
corresponding offer.
Note to entry: A dialogue act, A2, may also depend on another dialogue act, A1, occurring earlier in a dialogue because
of relations between their semantic contents, e.g. because A2 contains a reference to an element occurring in A1. This is
not a functional dependence relation, since it is not due to A2's communicative function.
3.11
functional segment
minimal stretch of communicative behaviour that has one or more communicative functions (3.4)
EXAMPLE The functional segment corresponding to the answer given by S in the following dialogue fragment does
not include the parts “Just a moment please” and “. let me see.” but only the parts “the first train to the airport on
Sunday morning is” and “at 5:45”:
1. U: What time is the first train to the airport on Sunday morning please?
2. S: Just a moment please. the first train to the airport on Sunday morning is . let me see. at 5:45.
Note 1 to entry: A consequence of this definition is that functional segments may be discontinuous, may overlap or be
embedded and may contain parts contributed by different participants.
Note 2 to entry: The condition of being “minimal” ensures that functional segments do not include material that does
not contribute to the expression of a communicative function that identifies the segment.
3.12
information state
context
totality of a dialogue (3.5) participant's (3.13) beliefs, assumptions, expectations, goals, preferences, hopes
and other attitudes that may influence the participant's interpretation and generation of communicative
behaviour
3.13
participant
person or artificial agent involved in the exchange of utterances (3.22)
3.14
qualifier
predicate that can be associated with a communicative function (3.4)
EXAMPLE A: Would you like to have some coffee?
B: Only if you have it ready.
B's utterance accepts A's offer under a certain condition; this can be described by qualifying the communicative function
Accept Offer with the predicate “conditional”.
Note to entry: See 10.3 for more examples.
3.15
rhetorical relation
relation between two dialogue acts (3.6), indicating a pragmatic connection between the two or between their
semantic contents (3.16)
EXAMPLE 1 The statement in the second utterance which follows provides a motivation for the question in the first
utterance:
A: Can you tell me what flights there are to Sydney on Saturday? I'd like to attend my mother's 80th birthday.
EXAMPLE 2 A rhetorical relation between the semantic contents of two dialogue act occurs in the following, where the
content of B's statement mentions a cause for the content of A's statement:
A: I can never find these stupid remote controls
B: That's because they don't have a fixed location
Note to entry: Relations such as elaboration, explanation, justification, cause and concession have been studied
extensively in the analysis of (monologue) text, where they are often called “rhetorical relations” or “discourse relations”
and are mostly viewed either as relations between text segments or as relations between events or propositions,
described in text segments. See, for example, Hovy and Maier, 1992, Lascarides & Asher, 2007 or Mann & Thompson,
1988. Many of these relations also occur in dialogue, either as relations between dialogue acts or between the semantic
contents of dialogue acts.
3.16
semantic content
information, situation, action, event or objects that a stretch of communicative behaviour refers to
3.17
semantic content category
semantic content type
kind of information, situation, action, event or objects that form the semantic content (3.16) of a dialogue act
(3.6)
EXAMPLE The various dimensions (3.7) defined in this part of ISO 24617 correspond to categories of semantic
content. In particular, the Task dimension corresponds to the category of task-specific actions and information; the Allo-
and Auto-Feedback dimensions correspond to the categories of information about the processing by the current speaker
or by the addressee, respectively, of something that was said before; the Turn Management dimension corresponds to the
category of information about the allocation of the speaker role and so forth.
3.18
sender
dialogue (3.5) participant (3.13) who produces a dialogue act (3.6)
3.19
speaker
sender (3.18) of a dialogue act (3.6) in the form of speech, possibly combined with nonverbal communicative
behaviour
Note to entry: A dialogue participant may say something while another participant occupies the speaker role (3.20),
therefore the term “speaker” is not synonymous with “participant who occupies the speaker role”.
4 © ISO 2012 – All rights reserved

3.20
speaker role
role occupied by a dialogue (3.5) participant (3.13) who has temporary control of the dialogue and speaks for
some period of time
[SOURCE: DAMSL Revised Manual.]
3.21
turn unit
stretch of communicative activity produced by one participant (3.13) who occupies the speaker role (3.20),
bounded by periods where another participant occupies the speaker role
3.22
utterance
anything said, written, keyed, gesticulated or otherwise expressed
Note to entry: An utterance is mostly a part of what a sender contributes in a turn unit.
4 Purpose and justification
The notion of a dialogue act plays a key role in the analysis of spoken and multimodal dialogue, as well as in
the design of spoken dialogue systems and embodied conversational agents. These activities all depend on
the availability of dialogue corpora, annotated with dialogue act information.
Over the years a variety of dialogue act annotation schemes have been developed, such as those of the
TRAINS human-computer dialogue project (Allen et al., 1994), the Map Task studies of human-human
dialogue (Carletta et al., 1996) and of the Verbmobil speech translation project (Alexandersson et al., 1998).
These schemes were developed for specific purposes and application domains. They contain overlapping sets
of concepts and make use of often mutually inconsistent terminology, sometimes employing different terms for
the same concept or the same term for different concepts.
The multidimensional DIT scheme (Bunt, 1984) was developed for information-seeking dialogues without
depending on a particular domain. The DAMSL scheme (Dialogue Act Markup using Several Layers, Allen
and Core,1997; Core et al., 1998) constitutes an application-independent multidimensional annotation
++
scheme. The DIT scheme (Bunt, 2006; 2009) combines the DIT scheme with concepts from DAMSL and
other more recent schemes into a comprehensive general-purpose annotation scheme.
In the EU-funded project LIRICS (Linguistic Infrastructure for Interoperable Resources and Systems, Romary
++
et al., 2007) a reference set of dialogue acts, taken from the DIT taxonomy, was defined in the form of data
categories, following ISO 12620. This set of concepts has been tested for its usability and coverage a) in the
manual annotation of spoken dialogues in English, Dutch and Italian and b) in the automatic annotation of
spoken and multimodal dialogue in English and forms a significant part of the background of this part of
ISO 24617.
The main purpose of this part of ISO 24617 is to define a reference set of domain-independent basic concepts
for dialogue act annotation, plus a formal language, based on XML, for representing such annotations.
Guidelines are provided for how to use the defined concepts and the annotation language, supported by
extended examples. This formal language, the Dialogue act markup language (DiAML) has a formal
semantics, which makes it possible to apply techniques for automatic reasoning to DiAML annotations.
Guidelines and principles are also provided for extending the set of concepts defined in this part of ISO 24617,
for example, with domain-specific concepts, as well as for selecting coherent subsets.
5 Basic concepts and metamodel
The term “dialogue act” is often used rather loosely in the sense of a speech act used in dialogue. Indeed, the
idea of interpreting communicative behaviour in terms of actions, such as questions, promises and requests,
goes back to speech act theory (Austin, 1962; Searle, 1969). But where speech act theory is primarily an
action-based approach to meaning within the philosophy of language, dialogue act theory is an
empirically-based approach to the computational modelling of linguistic and nonverbal communicative
behaviour in dialogue.
Dialogue acts offer a way of characterizing the meaning of communicative behaviour in terms of update
operations, to be applied to the information states of participants in the dialogue; this approach is commonly
known as the “information-state update” or “context-change” approach — see e.g. Bunt (1989; 2000a); Traum
and Larsson (2003). For instance, when an addressee understands the utterance “Do you know what time it
is?” as a question about the time, then the addressee's information state is updated to contain (among other
things) the information that the speaker does not know what time it is and would like to know that. If, by
contrast, it is understood that the speaker is reproaching the addressee for being late, then the addressee's
information state is updated to include (among other things) the information that the speaker does know what
time it is. Distinctions such as that between a question and a reproach concern the communicative function of
a dialogue act, which is one of its two main components. The other main component is its semantic content,
which describes the objects, properties, relations, situations, actions or events that the dialogue act is about.
The communicative function of a dialogue act specifies how an addressee should update his information state
with the information expressed in the semantic content when he understands the dialogue act.
A dialogue act as defined in this part of ISO 24617 (3.6) is a semantic unit of communicative behaviour.
Dialogue act annotation is the marking up of stretches of dialogue with information about the dialogue acts
performed in these stretches and is often limited to assigning communicative function tags. A dialogue act
being a semantic unit in communicative behaviour, the question arises as to which stretches of communicative
behaviour are considered as corresponding to dialogue acts. Spoken dialogues are traditionally segmented
into turns, defined as stretches of communicative behaviour produced by one speaker, bounded by periods of
inactivity of that speaker. Turns in this sense can be quite long and complex and are therefore not very useful
units of behaviour for assigning communicative functions to. Communicative functions can be assigned more
accurately to smaller units, which are called functional segments and which are defined as the minimal
stretches of communicative behaviour that are functionally relevant. See Clause 8 for more details about
dialogue segmentation.
Inherent to the notion of a dialogue act is that there is an agent who produces the dialogue act, called the
“sender” and one or more agents who are addressed, called “addressees”. Dialogue studies often focus on
two-person dialogues, in which case the dialogue acts have only one addressee. Besides sender and
addressee(s), there may be various types of side-participants who are present but do not or only marginally
participate (see Clark, 1996).
Dialogue act annotation is often limited to assigning communicative functions to dialogue segments, which
corresponds intuitively to indicating the type of communicative action that is performed. A semantically more
complete characterization also provides information about the type of semantic content. The DAMSL
annotation scheme distinguishes three categories of semantic content: task, task management and
communication, which indicate whether the semantic content of the dialogue act is concerned with performing
the task which underlies the dialogue or with discussing how to perform the task or with the communication.
++
The DIT scheme distinguishes a number of subcategories of communication-related information, such as
feedback information, turn allocation information and topic progression information. The various categories of
semantic content are also called “dimensions” and are discussed in more detail in Clause 7.
Some types of dialogue acts are inherently dependent for their full meaning on one or more dialogue acts that
occurred earlier in the dialogue. This is, for example, the case for answers, whose meaning is partly
determined by the question being answered and for the acceptance or rejection of offers, suggestions,
requests and apologies. The following example illustrates this, where the meaning of (1.3) clearly depends
very much on whether it is an answer to the question (1.1) or to the question (1.2).
EXAMPLE 1
(1.1) B: Do you know who's coming tonight?
6 © ISO 2012 – All rights reserved

(1.2) B: Which of the project members d'you think will be there?
(1.3) A: I'm expecting Jan, Alex, Claudia and David, and maybe Olga and Andrei.
As an answer to (1.1), it says that nobody else is expected to come than the people that are mentioned, but as
an answer to (1.2) it leaves open the possibility that other people will come, who are not members of “the
project”.
For dialogue acts which have such a dependence on other dialogue acts, due to their responsive character,
the marking up of the links to these “antecedent” dialogue acts allows the annotation not just to express e.g.
that the utterance is an answer, but also to express to which question it is an answer. This type of relation
between dialogue acts is called a functional dependence relation.
Dialogue acts may also be semantically related through other relations, as shown in the following example:
EXAMPLE 2
(2.1) A: It ties you on in terms of the technology and the complexity that you want
(2.2) A: like for example voice recognition
(2.3) A: because you might need to power a microphone and other things
(2.4) A: so that's one constraint there
2)
In this example we see a sequence of four functional segments contributed by the same participant.
Segment (2.2) is related to the initial statement through an Exemplification relation and (2.3) through an
Explanation relation, while (2.4) is related to the preceding three segments through a Summarization relation.
Such relations are known as rhetorical relations. In view of the wide diversity of the sets of rhetorical relations
that have been proposed (see, e.g., Mann and Thompson, 1988; Hovy and Maier, 1993; Sanders et al., 1992),
this part of ISO 24617 does not propose any specific set of such relations, but only provides a conceptual
category for which a particular set of relations may be specified.
Feedback-providing and eliciting acts also relate to what happened earlier in the dialogue, but in a different
way. They are concerned with the processing of what was said before — such as its perception or its
interpretation:
EXAMPLE 3
(3.1) A: Is this flight also available on Thursday?
(3.2a) B: On Thursday you said?
(3.2b) B: The twelfth you mean?
With utterance (3.2a), B checks whether he heard correctly what A said. This is a response to A's utterance,
rather than to the dialogue act that the utterance expresses; with utterance (3.2b), by contrast, B checks
whether he has correctly interpreted what A said. Both types of dependence are called a feedback
dependence relation.
Note that nonverbal feedback, for instance in the form of nodding or vocal backchannels like “uh-huh”, “um”,
“huh”, “m-hm”, may have a feedback dependence relation to what is being said at that moment, rather than to
what was said before. This is also the case for speech editing acts like self-corrections (“on Tuesday I mean
Thursday”) and completions of what the partner is trying to say.
Example 1 above also illustrates another phenomenon that is frequently found in dialogue, namely that
speakers may have incomplete or uncertain information. The use of “maybe” in (1.3) expresses that A is
uncertain about part of the information that he provides.

2) From the AMI corpus, see http://corpus.amiproject.org.
In addition, speakers may express a certain sentiment about the information or event that is being discussed,
as in (4.2) or express a reservation in the form of a condition, as in (4.3), where an offer is conditionally
accepted:
EXAMPLE 4
(4.1) A: Would you like to have some coffee?
(4.2) B: That would be great, thank you!
(4.3) B: Only if you have it ready.
For the annotation of conditions, uncertainty and sentiment, this part of ISO 24617 makes use of so-called
function qualifiers, which can be attached to communicative functions — see 10.3 for more detail.
The above characterization of the notion of a dialogue act makes use of the following key concepts, which
form the backbone of the metamodel for dialogue act annotation in Figure 1:
a) sender, addressee and participants in other roles (side-participants);
b) functional segment;
c) dialogue act, communicative function, communicative function qualifier and semantic content category (or
“dimension”);
d) functional dependence relation, rhetorical relation and feedback dependence relation.

Figure 1 — Metamodel for dialogue act annotation
6 Definition of communicative functions
Existing dialogue act annotation schemes use one of the following two approaches to defining communicative
functions or a combination of the two: (1) in terms of the effects on addressees intended by the sender; (2) in
terms of properties of the signals that are used. Defining a communicative function by its linguistic form has
the advantage that its recognition can be straightforward, but runs into the problem that the same linguistic
form can be used to express different functions. For example, the utterance “Why don't you start?” has the
form of a question and can be intended as such, but can also be used to invite or encourage somebody to
8 © ISO 2012 – All rights reserved

start. Similarly for so-called “declarative questions” (questions in the form of a declarative sentence), like
“You're going home tomorrow”, which are intended as questions although they look like statements.
Form-based definitions also run the risk of being purely descriptive, rather than semantic. For example, when
a speaker repeats something that was said before, this behaviour may be characterized as a repetition;
however, that would only say something about the form of the behaviour, nothing about its communicative
function. A repetition for instance often has a feedback function, as in (5.2a) but it can also have other
functions, as in (5.3), where it is used as a confirmation in response to a check question:
EXAMPLE 5
(5.1) S: There are evening flights at seven-fifteen and eight-thirty
(5.2a) C: Seven-fifteen and eight-thirty
(5.2b) C: And that's on Sunday too
(5.3) S: And that's on Sunday too
This part of ISO 24617 follows a strictly semantic approach to the definition of communicative functions. But
while linguistic form is taken not to be part of the definition of a communicative function, a requirement for
introducing a communicative function is that there are ways in which a sender can indicate that his behaviour
should be understood as having that particular function, by shaping his (linguistic and/or nonverbal) behaviour
so as to have certain observable features which are indicative for that function in the context in which the
behaviour occurs. This requirement puts all communicative functions on an empirical basis.
A particular case where form and function are not related in a straightforward way is that of indirect speech
acts, where a speaker uses a linguistic form that is standardly used to express one type of dialogue act, but in
context means something else. Questions of the form Do you know [X] are illustrative: while an utterance of
this form would standardly seem to ask an addressee whether he possess the knowledge [X], it is more often
used to request the addressee to provide the information [X], if possible. This makes such a question a
conditional request.
The full complexity of the phenomenon of indirect speech acts is beyond the scope of this part of ISO 24617,
but an important class of indirect speech acts can be covered by qualifying them as conditional — see 10.3.
7 Annotation schemes
7.1 Structure of annotation schemes
Existing dialogue act annotation schemes can be divided into one-dimensional and multidimensional
schemes. One-dimensional schemes have a set of mutually exclusive tags and are used for coding stretches
of dialogue with a single tag. Multidimensional schemes, on the other hand, are intended for encoding
stretches of dialogue with multiple tags. Schemes of the latter kind typically have a relatively large tag set.
There are several advantages to the structuring of such a tag set into clusters of communicative functions
tags:
 Clustering semantically related tags improves the transparency of the tag set, as each cluster is
concerned with a certain kind of information. This also makes the coverage of the tag set clearer, since
each cluster typically corresponds to a certain class of dialogue phenomena.
 A structured tag set can be searched more systematically and more “semantically” (i.e. on the basis of
semantic differences and similarities) than an unstructured one.
 The tags within a cluster are usually mutually exclusive; this has the advantage that, once a particular tag
has been assigned, the rest of the tags within that cluster do not need to considered any further. If a
cluster is hierarchically organized, as is the case in this part of ISO 24617, with finer-grained functions
being dominated by less fine-grained ones (such as “confirmation” being more fine-grained than
“answer”), then the most sensible use of these tags is to choose the most specific tag for which there is
sufficient evidence.
7.2 Multidimensionality and multifunctionality
Participation in a dialogue involves several activities beyond those strictly related to performing the task or
activity for which the dialogue is instrumental. In natural conversation, the participants among other things
constantly “evaluate whether and how they can (and/or wish to) continue, perceive, understand and react to
each other's intentions” (Allwood, 1997). Communication is thus a complex, multi-faceted activity and this is
reflected in the multifunctionality that dialogue utterances often exhibit.
Multifunctionality comes in a variety of forms. Allwood (1992) distinguishes between sequential and
simultaneous multifunctionality and provides the following example as an illustration:
EXAMPLE 6 A: Yes! Come tomorrow. Go to the church. Bill will be there. OK?
B: The church, OK.
Sequential multifunctionality occurs when a turn has several parts which each have a different communicative
function. In Example 6 we see A's utterance containing five functional segments, with communicative
functions such as feedback giving, request, request, statement and response elicitation The occurrence of
sequential multifunctionality depends on the way in which a dialogue is segmented (see also Clause 8) and
disappears when sufficiently small segments are considered as markables.
Simultaneous multifunctionality, by contrast, persists even when minimal segments are used as markables.
The following example illustrates this:
EXAMPLE 7
(7.1) A: Do you know what date it is?
(7.2) B: Today is the fifteenth.
(7.3) A: Thank you.
A's utterance (7.3) has the function of thanking and will mostly be taken to imply that A has understood and
accepted the information in (7.2) — i.e. as having a positive feedback function. But “Thank you” does not
always express positive feedback; a participant in an unsuccessful dialogue may just want to terminate the
interaction in a polite way. The feedback function of the thanking in (7.3) can be inferred along the following
lines: By saying “Thank you”, A expresses his gratitude to B. This can only be for what B just said; this would
constitute a reason for being grateful if A considers B's utterance as relevant and useful, which means that A
accepted B's utterance as an answer to his question. The feedback function in such a case can be viewed as
a conversational implicature (Grice, 1979), i.e. as a contextually plausible consequence which the addressee
is intended to infer.
The implication relation between thanking and positive feedback is different from that between a propositional
answer (“yes” or “no”) and a confirmation, where the relation is one of entailment, i.e. an implication which is
logically valid. (Every confirmation by its very nature is also an answer.) Entailment relations occur when the
definition of one communicative function is a special case of that of another.
It may be argued that such cases should not be considered as instances of multifunctionality, e.g. a speaker
who wants to issue a confirmation can hardly have the intention of additionally giving an answer, since the
recognition of that intention is already part of the recognition of a confirmation.
There are also cases of multifunctionality where the different functions do not have any logical relation. This is,
for example, the case for turn-initial hesitations, as in the following dialogue fragment:
EXAMPLE 8
(8.1) A: Is that your opinion too?
10 © ISO 2012 – All rights reserved

(8.2) B: Uh,. well,. I guess so.
In (8.1), speaker A asks a question to B and assigns the turn to B. In (8.2) B performs a stalling act in order to
buy some time for deciding what to say; the fact that he starts speaking without waiting until he has made up
his mind about what to say, indicates that he accepts the turn. So the segment “Uh,. well,.” is multifunctional,
having both a stalling function and a turn-accepting function. Note that A's utterance is also multifunctional: it
asks a question about B's opinion and it assigns the turn to B (due to its intonation, in combination with A
looking at B and raising his eyebrows).
The design of a dialogue act annotation schema can reflect the multifunctional view of utterances in two ways:
1) by structuring the tag set into clusters (see below); 2) by accompanying instructions to annotators for how
to apply multiple tags. If the tag set is fairly extended and does n
...

INTERNATIONAL ISO
STANDARD 24617-2
First edition
2012-09-01
Language resource management —
Semantic annotation framework
(SemAF) —
Part 2:
Dialogue acts
Gestion des ressources langagières — Cadre d'annotation sémantique
(SemAF) —
Partie 2: Actes de dialogue
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 . iv
1  Scope . 1
2  Normative references . 1
3  Terms and definitions . 1
4  Purpose and justification . 5
5  Basic concepts and metamodel . 6
6  Definition of communicative functions . 8
7  Annotation schemes . 9
7.1  Structure of annotation schemes . 9
7.2  Multidimensionality and multifunctionality . 10
7.3  Multidimensionality, clustering and dimensions . 11
7.4  Dimension- specific and general-purpose functions . 11
8  Dialogue segmentation . 13
9  Dimensions . 14
9.1  Task. 15
9.2  Auto-Feedback . 15
9.3  Allo-Feedback . 15
9.4  Turn Management . 15
9.5  Time Management . 16
9.6  Discourse Structuring . 16
9.7  Social Obligations Management . 16
9.8  Own Communication Management . 16
9.9  Partner Communication Management . 16
10  Core dialogue acts . 17
10.1  General-purpose functions . 19
10.2  Dimension-specific functions . 20
10.3  Function qualifiers . 22
11  Dialogue act markup language (DiAML) . 23
11.1  Abstract syntax . 23
11.2  Concrete syntax . 24
12  Principles for extending and restricting the standard . 25
12.1  Main design principles . 25
12.2  Schema extension . 27
12.3  Scheme restriction . 27
Annex A (informative) Annotation guidelines . 29
Annex B (informative) Annotated dialogue examples . 43
Annex C (normative) Formal definition of DiAML . 56
Annex D (normative) DiAML technical schema . 63
Annex E (normative)  Data categories for core concepts . 68
Annex F (informative) Examples of possible additional data categories . 88
Annex G (informative) Concepts in existing schemes . 90
Bibliography . 100
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 24617-2 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:
 Part 1: Time and events (SemAF-Time, ISO-TimeML)
 Part 2: Dialogue acts
The following parts are under preparation:
 Part 3: Named entities (SemAF-NE)
 Part 4: Semantic roles (SemAF-SRL)
 Part 5: Discourse structure (SemAF-DS)
 Part 6: Principles of semantic annotation (SemAF-Basics)
 Part 7: Spatial information (ISO-Space)
 Part 8: Semantic relations in discourse (SemAF-DRel)

iv © ISO 2012 – All rights reserved

INTERNATIONAL STANDARD ISO 24617-2:2012(E)

Language resource management — Semantic annotation
framework (SemAF) —
Part 2:
Dialogue acts
1 Scope
This part of ISO 24617 provides a set of empirically and theoretically well-motivated concepts for dialogue
annotation, a formal language for expressing dialogue annotations — the dialogue act markup language
(DiAML) — and a method for segmenting a dialogue into semantic units. This allows the manual or automatic
annotation of dialogue segments with information about the communicative actions which the participants
perform by their contributions to the dialogue. It supports multidimensional annotation, in which units in
dialogue are viewed as having multiple communicative functions. The DiAML language has an XML-based
representation format and a formal semantics which makes it possible to apply inference to DiAML
representations.
This part of ISO 24617 specifies data categories for reference sets of communicative functions and
dimensions of dialogue analysis and provides principles and guidelines for extending these sets or selecting
coherent subsets of them. Additionally, it provides guidelines for annotators and annotated examples. It is
applicable to spoken, written and multimodal dialogues involving two or more participants.
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.
ISO 12620:2009, Terminology and other language resources — Specification of data categories and
management of a Data Category Registry for language resources
ISO 24610-1:2006, Language resource management — Feature structures — Part 1: Feature structure
representation
ISO 24612:2011, Language resource management — Linguistic annotation framework
3 Terms and definitions
1)
For the purposes of this document, the following terms and definitions apply.

1) In this document, “he”, “him” and “his” are used in a generic sense, without implying any gender-related distinctions.
3.1
addressee
dialogue (3.5) participant (3.13) oriented to by the sender (3.18) in a manner to suggest that his utterances
(3.22) are particularly intended for this participant and that some response is therefore anticipated from this
participant, more so than from the other participants
Note to entry: This definition is a de facto standard in the linguistics literature. It has been slightly modified here, in
replacing “speaker” by “sender” and avoiding the use of ambiguous pronouns. Goffman's original definition says: “dialogue
participant oriented to by the speaker in a manner to suggest that his utterances are particularly intended for him and that
some response is therefore anticipated from him/her, more so than from the other participants”.
[SOURCE: Goffman (1981).]
3.2
allo-feedback act
feedback act (3.8) where the sender (3.18) elicits information about the addressee's (3.1) processing of an
utterance (3.22) that the sender contributed to the dialogue (3.5) or where the sender provides information
about his perceived processing by the addressee of an utterance that the sender contributed to the dialogue
before
EXAMPLE A: Now move up.
B: Slightly northeast you mean?
A: Slightly yeah.
A performs an allo-feedback act signalling that he thinks B understood his first utterance correctly.
3.3
auto-feedback act
feedback act (3.8) where the sender (3.18) provides information about his own processing of an utterance
(3.22) contributed to the dialogue (3.5) by another participant (3.13)
EXAMPLE B's utterance in the example dialogue fragment in (3.2) signals that he is uncertain whether he
understood the previous utterance correctly.
3.4
communicative function
property of certain stretches of communicative behaviour, describing how the behaviour changes the
information state (3.12) of an understander of the behaviour
Note to entry: A communicative function may be “qualified”, i.e. one or more qualifiers (3.14) may be associated with it.
For example, an answer may be qualified as “uncertain” and the acceptance of a request may be “conditional”. See 10.3
for explanation and examples.
3.5
dialogue
exchange of utterances (3.22) between two or more persons or artificial conversational systems
3.6
dialogue act
communicative activity of a dialogue (3.5) participant (3.13), interpreted as having a certain communicative
function (3.4) and semantic content (3.16)
Note to entry: A dialogue act may also have certain functional dependence relations (3.10), rhetorical relations (3.15) and
feedback dependence relations (3.9) with other units in a dialogue (3.5).
3.7
dimension
class of dialogue acts (3.6) that are concerned with a particular aspect of communication, corresponding to a
particular category of semantic content
2 © ISO 2012 – All rights reserved

EXAMPLE Dialogue acts advancing the task or activity that motivates the dialogue (the Task dimension), dialogue
acts providing and eliciting feedback (the Auto- and Allo-Feedback dimensions) and dialogue acts for allocating the
speaker role (the Turn Management dimension).
Note to entry: See Clauses 5, 7 and 9 for discussion and more examples.
3.8
feedback act
dialogue act (3.6) which provides or elicits information about the sender's (3.18) or the addressee's (3.1)
processing of something that was uttered in the dialogue
Note to entry: Two classes of feedback are distinguished in this part of ISO 24617: allo-feedback acts (3.2) and auto-
feedback acts (3.3).
3.9
feedback dependence relation
relation between a feedback act (3.8) and the stretch of communicative behaviour whose processing the act
provides or elicits information about
EXAMPLE In the example that accompanies definition 3.2, both the allo-feedback act expressed by utterance 3 and
the auto-feedback act expressed by utterance 2 have a feedback dependence relation to utterance 1.
3.10
functional dependence relation
relation between a given dialogue act (3.6) and a preceding dialogue act on which the semantic content of the
given dialogue act depends due to its communicative function (3.4)
EXAMPLE The relation between an answer and the corresponding question, such as between utterance 3 and
utterance 2 in the example accompanying definition 3.2; or the relation between the acceptance of an offer and the
corresponding offer.
Note to entry: A dialogue act, A2, may also depend on another dialogue act, A1, occurring earlier in a dialogue because
of relations between their semantic contents, e.g. because A2 contains a reference to an element occurring in A1. This is
not a functional dependence relation, since it is not due to A2's communicative function.
3.11
functional segment
minimal stretch of communicative behaviour that has one or more communicative functions (3.4)
EXAMPLE The functional segment corresponding to the answer given by S in the following dialogue fragment does
not include the parts “Just a moment please” and “. let me see.” but only the parts “the first train to the airport on
Sunday morning is” and “at 5:45”:
1. U: What time is the first train to the airport on Sunday morning please?
2. S: Just a moment please. the first train to the airport on Sunday morning is . let me see. at 5:45.
Note 1 to entry: A consequence of this definition is that functional segments may be discontinuous, may overlap or be
embedded and may contain parts contributed by different participants.
Note 2 to entry: The condition of being “minimal” ensures that functional segments do not include material that does
not contribute to the expression of a communicative function that identifies the segment.
3.12
information state
context
totality of a dialogue (3.5) participant's (3.13) beliefs, assumptions, expectations, goals, preferences, hopes
and other attitudes that may influence the participant's interpretation and generation of communicative
behaviour
3.13
participant
person or artificial agent involved in the exchange of utterances (3.22)
3.14
qualifier
predicate that can be associated with a communicative function (3.4)
EXAMPLE A: Would you like to have some coffee?
B: Only if you have it ready.
B's utterance accepts A's offer under a certain condition; this can be described by qualifying the communicative function
Accept Offer with the predicate “conditional”.
Note to entry: See 10.3 for more examples.
3.15
rhetorical relation
relation between two dialogue acts (3.6), indicating a pragmatic connection between the two or between their
semantic contents (3.16)
EXAMPLE 1 The statement in the second utterance which follows provides a motivation for the question in the first
utterance:
A: Can you tell me what flights there are to Sydney on Saturday? I'd like to attend my mother's 80th birthday.
EXAMPLE 2 A rhetorical relation between the semantic contents of two dialogue act occurs in the following, where the
content of B's statement mentions a cause for the content of A's statement:
A: I can never find these stupid remote controls
B: That's because they don't have a fixed location
Note to entry: Relations such as elaboration, explanation, justification, cause and concession have been studied
extensively in the analysis of (monologue) text, where they are often called “rhetorical relations” or “discourse relations”
and are mostly viewed either as relations between text segments or as relations between events or propositions,
described in text segments. See, for example, Hovy and Maier, 1992, Lascarides & Asher, 2007 or Mann & Thompson,
1988. Many of these relations also occur in dialogue, either as relations between dialogue acts or between the semantic
contents of dialogue acts.
3.16
semantic content
information, situation, action, event or objects that a stretch of communicative behaviour refers to
3.17
semantic content category
semantic content type
kind of information, situation, action, event or objects that form the semantic content (3.16) of a dialogue act
(3.6)
EXAMPLE The various dimensions (3.7) defined in this part of ISO 24617 correspond to categories of semantic
content. In particular, the Task dimension corresponds to the category of task-specific actions and information; the Allo-
and Auto-Feedback dimensions correspond to the categories of information about the processing by the current speaker
or by the addressee, respectively, of something that was said before; the Turn Management dimension corresponds to the
category of information about the allocation of the speaker role and so forth.
3.18
sender
dialogue (3.5) participant (3.13) who produces a dialogue act (3.6)
3.19
speaker
sender (3.18) of a dialogue act (3.6) in the form of speech, possibly combined with nonverbal communicative
behaviour
Note to entry: A dialogue participant may say something while another participant occupies the speaker role (3.20),
therefore the term “speaker” is not synonymous with “participant who occupies the speaker role”.
4 © ISO 2012 – All rights reserved

3.20
speaker role
role occupied by a dialogue (3.5) participant (3.13) who has temporary control of the dialogue and speaks for
some period of time
[SOURCE: DAMSL Revised Manual.]
3.21
turn unit
stretch of communicative activity produced by one participant (3.13) who occupies the speaker role (3.20),
bounded by periods where another participant occupies the speaker role
3.22
utterance
anything said, written, keyed, gesticulated or otherwise expressed
Note to entry: An utterance is mostly a part of what a sender contributes in a turn unit.
4 Purpose and justification
The notion of a dialogue act plays a key role in the analysis of spoken and multimodal dialogue, as well as in
the design of spoken dialogue systems and embodied conversational agents. These activities all depend on
the availability of dialogue corpora, annotated with dialogue act information.
Over the years a variety of dialogue act annotation schemes have been developed, such as those of the
TRAINS human-computer dialogue project (Allen et al., 1994), the Map Task studies of human-human
dialogue (Carletta et al., 1996) and of the Verbmobil speech translation project (Alexandersson et al., 1998).
These schemes were developed for specific purposes and application domains. They contain overlapping sets
of concepts and make use of often mutually inconsistent terminology, sometimes employing different terms for
the same concept or the same term for different concepts.
The multidimensional DIT scheme (Bunt, 1984) was developed for information-seeking dialogues without
depending on a particular domain. The DAMSL scheme (Dialogue Act Markup using Several Layers, Allen
and Core,1997; Core et al., 1998) constitutes an application-independent multidimensional annotation
++
scheme. The DIT scheme (Bunt, 2006; 2009) combines the DIT scheme with concepts from DAMSL and
other more recent schemes into a comprehensive general-purpose annotation scheme.
In the EU-funded project LIRICS (Linguistic Infrastructure for Interoperable Resources and Systems, Romary
++
et al., 2007) a reference set of dialogue acts, taken from the DIT taxonomy, was defined in the form of data
categories, following ISO 12620. This set of concepts has been tested for its usability and coverage a) in the
manual annotation of spoken dialogues in English, Dutch and Italian and b) in the automatic annotation of
spoken and multimodal dialogue in English and forms a significant part of the background of this part of
ISO 24617.
The main purpose of this part of ISO 24617 is to define a reference set of domain-independent basic concepts
for dialogue act annotation, plus a formal language, based on XML, for representing such annotations.
Guidelines are provided for how to use the defined concepts and the annotation language, supported by
extended examples. This formal language, the Dialogue act markup language (DiAML) has a formal
semantics, which makes it possible to apply techniques for automatic reasoning to DiAML annotations.
Guidelines and principles are also provided for extending the set of concepts defined in this part of ISO 24617,
for example, with domain-specific concepts, as well as for selecting coherent subsets.
5 Basic concepts and metamodel
The term “dialogue act” is often used rather loosely in the sense of a speech act used in dialogue. Indeed, the
idea of interpreting communicative behaviour in terms of actions, such as questions, promises and requests,
goes back to speech act theory (Austin, 1962; Searle, 1969). But where speech act theory is primarily an
action-based approach to meaning within the philosophy of language, dialogue act theory is an
empirically-based approach to the computational modelling of linguistic and nonverbal communicative
behaviour in dialogue.
Dialogue acts offer a way of characterizing the meaning of communicative behaviour in terms of update
operations, to be applied to the information states of participants in the dialogue; this approach is commonly
known as the “information-state update” or “context-change” approach — see e.g. Bunt (1989; 2000a); Traum
and Larsson (2003). For instance, when an addressee understands the utterance “Do you know what time it
is?” as a question about the time, then the addressee's information state is updated to contain (among other
things) the information that the speaker does not know what time it is and would like to know that. If, by
contrast, it is understood that the speaker is reproaching the addressee for being late, then the addressee's
information state is updated to include (among other things) the information that the speaker does know what
time it is. Distinctions such as that between a question and a reproach concern the communicative function of
a dialogue act, which is one of its two main components. The other main component is its semantic content,
which describes the objects, properties, relations, situations, actions or events that the dialogue act is about.
The communicative function of a dialogue act specifies how an addressee should update his information state
with the information expressed in the semantic content when he understands the dialogue act.
A dialogue act as defined in this part of ISO 24617 (3.6) is a semantic unit of communicative behaviour.
Dialogue act annotation is the marking up of stretches of dialogue with information about the dialogue acts
performed in these stretches and is often limited to assigning communicative function tags. A dialogue act
being a semantic unit in communicative behaviour, the question arises as to which stretches of communicative
behaviour are considered as corresponding to dialogue acts. Spoken dialogues are traditionally segmented
into turns, defined as stretches of communicative behaviour produced by one speaker, bounded by periods of
inactivity of that speaker. Turns in this sense can be quite long and complex and are therefore not very useful
units of behaviour for assigning communicative functions to. Communicative functions can be assigned more
accurately to smaller units, which are called functional segments and which are defined as the minimal
stretches of communicative behaviour that are functionally relevant. See Clause 8 for more details about
dialogue segmentation.
Inherent to the notion of a dialogue act is that there is an agent who produces the dialogue act, called the
“sender” and one or more agents who are addressed, called “addressees”. Dialogue studies often focus on
two-person dialogues, in which case the dialogue acts have only one addressee. Besides sender and
addressee(s), there may be various types of side-participants who are present but do not or only marginally
participate (see Clark, 1996).
Dialogue act annotation is often limited to assigning communicative functions to dialogue segments, which
corresponds intuitively to indicating the type of communicative action that is performed. A semantically more
complete characterization also provides information about the type of semantic content. The DAMSL
annotation scheme distinguishes three categories of semantic content: task, task management and
communication, which indicate whether the semantic content of the dialogue act is concerned with performing
the task which underlies the dialogue or with discussing how to perform the task or with the communication.
++
The DIT scheme distinguishes a number of subcategories of communication-related information, such as
feedback information, turn allocation information and topic progression information. The various categories of
semantic content are also called “dimensions” and are discussed in more detail in Clause 7.
Some types of dialogue acts are inherently dependent for their full meaning on one or more dialogue acts that
occurred earlier in the dialogue. This is, for example, the case for answers, whose meaning is partly
determined by the question being answered and for the acceptance or rejection of offers, suggestions,
requests and apologies. The following example illustrates this, where the meaning of (1.3) clearly depends
very much on whether it is an answer to the question (1.1) or to the question (1.2).
EXAMPLE 1
(1.1) B: Do you know who's coming tonight?
6 © ISO 2012 – All rights reserved

(1.2) B: Which of the project members d'you think will be there?
(1.3) A: I'm expecting Jan, Alex, Claudia and David, and maybe Olga and Andrei.
As an answer to (1.1), it says that nobody else is expected to come than the people that are mentioned, but as
an answer to (1.2) it leaves open the possibility that other people will come, who are not members of “the
project”.
For dialogue acts which have such a dependence on other dialogue acts, due to their responsive character,
the marking up of the links to these “antecedent” dialogue acts allows the annotation not just to express e.g.
that the utterance is an answer, but also to express to which question it is an answer. This type of relation
between dialogue acts is called a functional dependence relation.
Dialogue acts may also be semantically related through other relations, as shown in the following example:
EXAMPLE 2
(2.1) A: It ties you on in terms of the technology and the complexity that you want
(2.2) A: like for example voice recognition
(2.3) A: because you might need to power a microphone and other things
(2.4) A: so that's one constraint there
2)
In this example we see a sequence of four functional segments contributed by the same participant.
Segment (2.2) is related to the initial statement through an Exemplification relation and (2.3) through an
Explanation relation, while (2.4) is related to the preceding three segments through a Summarization relation.
Such relations are known as rhetorical relations. In view of the wide diversity of the sets of rhetorical relations
that have been proposed (see, e.g., Mann and Thompson, 1988; Hovy and Maier, 1993; Sanders et al., 1992),
this part of ISO 24617 does not propose any specific set of such relations, but only provides a conceptual
category for which a particular set of relations may be specified.
Feedback-providing and eliciting acts also relate to what happened earlier in the dialogue, but in a different
way. They are concerned with the processing of what was said before — such as its perception or its
interpretation:
EXAMPLE 3
(3.1) A: Is this flight also available on Thursday?
(3.2a) B: On Thursday you said?
(3.2b) B: The twelfth you mean?
With utterance (3.2a), B checks whether he heard correctly what A said. This is a response to A's utterance,
rather than to the dialogue act that the utterance expresses; with utterance (3.2b), by contrast, B checks
whether he has correctly interpreted what A said. Both types of dependence are called a feedback
dependence relation.
Note that nonverbal feedback, for instance in the form of nodding or vocal backchannels like “uh-huh”, “um”,
“huh”, “m-hm”, may have a feedback dependence relation to what is being said at that moment, rather than to
what was said before. This is also the case for speech editing acts like self-corrections (“on Tuesday I mean
Thursday”) and completions of what the partner is trying to say.
Example 1 above also illustrates another phenomenon that is frequently found in dialogue, namely that
speakers may have incomplete or uncertain information. The use of “maybe” in (1.3) expresses that A is
uncertain about part of the information that he provides.

2) From the AMI corpus, see http://corpus.amiproject.org.
In addition, speakers may express a certain sentiment about the information or event that is being discussed,
as in (4.2) or express a reservation in the form of a condition, as in (4.3), where an offer is conditionally
accepted:
EXAMPLE 4
(4.1) A: Would you like to have some coffee?
(4.2) B: That would be great, thank you!
(4.3) B: Only if you have it ready.
For the annotation of conditions, uncertainty and sentiment, this part of ISO 24617 makes use of so-called
function qualifiers, which can be attached to communicative functions — see 10.3 for more detail.
The above characterization of the notion of a dialogue act makes use of the following key concepts, which
form the backbone of the metamodel for dialogue act annotation in Figure 1:
a) sender, addressee and participants in other roles (side-participants);
b) functional segment;
c) dialogue act, communicative function, communicative function qualifier and semantic content category (or
“dimension”);
d) functional dependence relation, rhetorical relation and feedback dependence relation.

Figure 1 — Metamodel for dialogue act annotation
6 Definition of communicative functions
Existing dialogue act annotation schemes use one of the following two approaches to defining communicative
functions or a combination of the two: (1) in terms of the effects on addressees intended by the sender; (2) in
terms of properties of the signals that are used. Defining a communicative function by its linguistic form has
the advantage that its recognition can be straightforward, but runs into the problem that the same linguistic
form can be used to express different functions. For example, the utterance “Why don't you start?” has the
form of a question and can be intended as such, but can also be used to invite or encourage somebody to
8 © ISO 2012 – All rights reserved

start. Similarly for so-called “declarative questions” (questions in the form of a declarative sentence), like
“You're going home tomorrow”, which are intended as questions although they look like statements.
Form-based definitions also run the risk of being purely descriptive, rather than semantic. For example, when
a speaker repeats something that was said before, this behaviour may be characterized as a repetition;
however, that would only say something about the form of the behaviour, nothing about its communicative
function. A repetition for instance often has a feedback function, as in (5.2a) but it can also have other
functions, as in (5.3), where it is used as a confirmation in response to a check question:
EXAMPLE 5
(5.1) S: There are evening flights at seven-fifteen and eight-thirty
(5.2a) C: Seven-fifteen and eight-thirty
(5.2b) C: And that's on Sunday too
(5.3) S: And that's on Sunday too
This part of ISO 24617 follows a strictly semantic approach to the definition of communicative functions. But
while linguistic form is taken not to be part of the definition of a communicative function, a requirement for
introducing a communicative function is that there are ways in which a sender can indicate that his behaviour
should be understood as having that particular function, by shaping his (linguistic and/or nonverbal) behaviour
so as to have certain observable features which are indicative for that function in the context in which the
behaviour occurs. This requirement puts all communicative functions on an empirical basis.
A particular case where form and function are not related in a straightforward way is that of indirect speech
acts, where a speaker uses a linguistic form that is standardly used to express one type of dialogue act, but in
context means something else. Questions of the form Do you know [X] are illustrative: while an utterance of
this form would standardly seem to ask an addressee whether he possess the knowledge [X], it is more often
used to request the addressee to provide the information [X], if possible. This makes such a question a
conditional request.
The full complexity of the phenomenon of indirect speech acts is beyond the scope of this part of ISO 24617,
but an important class of indirect speech acts can be covered by qualifying them as conditional — see 10.3.
7 Annotation schemes
7.1 Structure of annotation schemes
Existing dialogue act annotation schemes can be divided into one-dimensional and multidimensional
schemes. One-dimensional schemes have a set of mutually exclusive tags and are used for coding stretches
of dialogue with a single tag. Multidimensional schemes, on the other hand, are intended for encoding
stretches of dialogue with multiple tags. Schemes of the latter kind typically have a relatively large tag set.
There are several advantages to the structuring of such a tag set into clusters of communicative functions
tags:
 Clustering semantically related tags improves the transparency of the tag set, as each cluster is
concerned with a certain kind of information. This also makes the coverage of the tag set clearer, since
each cluster typically corresponds to a certain class of dialogue phenomena.
 A structured tag set can be searched more systematically and more “semantically” (i.e. on the basis of
semantic differences and similarities) than an unstructured one.
 The tags within a cluster are usually mutually exclusive; this has the advantage that, once a particular tag
has been assigned, the rest of the tags within that cluster do not need to considered any further. If a
cluster is hierarchically organized, as is the case in this part of ISO 24617, with finer-grained functions
being dominated by less fine-grained ones (such as “confirmation” being more fine-grained than
“answer”), then the most sensible use of these tags is to choose the most specific tag for which there is
sufficient evidence.
7.2 Multidimensionality and multifunctionality
Participation in a dialogue involves several activities beyond those strictly related to performing the task or
activity for which the dialogue is instrumental. In natural conversation, the participants among other things
constantly “evaluate whether and how they can (and/or wish to) continue, perceive, understand and react to
each other's intentions” (Allwood, 1997). Communication is thus a complex, multi-faceted activity and this is
reflected in the multifunctionality that dialogue utterances often exhibit.
Multifunctionality comes in a variety of forms. Allwood (1992) distinguishes between sequential and
simultaneous multifunctionality and provides the following example as an illustration:
EXAMPLE 6 A: Yes! Come tomorrow. Go to the church. Bill will be there. OK?
B: The church, OK.
Sequential multifunctionality occurs when a turn has several parts which each have a different communicative
function. In Example 6 we see A's utterance containing five functional segments, with communicative
functions such as feedback giving, request, request, statement and response elicitation The occurrence of
sequential multifunctionality depends on the way in which a dialogue is segmented (see also Clause 8) and
disappears when sufficiently small segments are considered as markables.
Simultaneous multifunctionality, by contrast, persists even when minimal segments are used as markables.
The following example illustrates this:
EXAMPLE 7
(7.1) A: Do you know what date it is?
(7.2) B: Today is the fifteenth.
(7.3) A: Thank you.
A's utterance (7.3) has the function of thanking and will mostly be taken to imply that A has understood and
accepted the information in (7.2) — i.e. as having a positive feedback function. But “Thank you” does not
always express positive feedback; a participant in an unsuccessful dialogue may just want to terminate the
interaction in a polite way. The feedback function of the thanking in (7.3) can be inferred along the following
lines: By saying “Thank you”, A expresses his gratitude to B. This can only be for what B just said; this would
constitute a reason for being grateful if A considers B's utterance as relevant and useful, which means that A
accepted B's utterance as an answer to his question. The feedback function in such a case can be viewed as
a conversational implicature (Grice, 1979), i.e. as a contextually plausible consequence which the addressee
is intended to infer.
The implication relation between thanking and positive feedback is different from that between a propositional
answer (“yes” or “no”) and a confirmation, where the relation is one of entailment, i.e. an implication which is
logically valid. (Every confirmation by its very nature is also an answer.) Entailment relations occur when the
definition of one communicative function is a special case of that of another.
It may be argued that such cases should not be considered as instances of multifunctionality, e.g. a speaker
who wants to issue a confirmation can hardly have the intention of additionally giving an answer, since the
recognition of that intention is already part of the recognition of a confirmation.
There are also cases of multifunctionality where the different functions do not have any logical relation. This is,
for example, the case for turn-initial hesitations, as in the following dialogue fragment:
EXAMPLE 8
(8.1) A: Is that your opinion too?
10 © ISO 2012 – All rights reserved

(8.2) B: Uh,. well,. I guess so.
In (8.1), speaker A asks a question to B and assigns the turn to B. In (8.2) B performs a stalling act in order to
buy some time for deciding what to say; the fact that he starts speaking without waiting until he has made up
his mind about what to say, indicates that he accepts the turn. So the segment “Uh,. well,.” is multifunctional,
having both a stalling function and a turn-accepting function. Note that A's utterance is also multifunctional: it
asks a question about B's opinion and it assigns the turn to B (due to its intonation, in combination with A
looking at B and raising his eyebrows).
The design of a dialogue act annotation schema can reflect the multifunctional view of utterances in two ways:
1) by structuring the tag set into clusters (see below); 2) by accompanying instructions to annotators for how
to apply multiple tags. If the tag set is fairly extended and does not have any structure, it is next to impossible
to formulate good instructions for how to apply multiple tags, since there is no easy way to refer to groups of
tags. Therefore, the recognition that utterances in dialogue tend to be multifunctional naturally leads to the
introduction of dimensions in a dialogue annotation schema.
7.3 Multidimensionality, clustering and dimensions
The clusters of communicative functions that can be found in existing annotation schemes are typically
chosen on the basis of a conceptual similarity of certain functions. An early version of the DIT schema, for
example, has a cluster of “information-seeking functions” for a range of question types and a cluster of
“information-providing” functions for various kinds of informs and answers (Bunt, 1989).
The DAMSL schema (Core and Allen, 1997) is organized into “layers” and “dimensions”. Four layers are
distinguished: communicative status, information level, and forward looking and backward looking
communicative functions (FLF and BLF); the latter two are indeed clusters of communicative functions (the
tags in the other layers are concerned with other kinds of information). The FLF cluster is subdivided into five
clusters, including the classes of commissive and directive functions, well known from s
...