[ Dialogue Macrogame Theory Home Page ]

Analyzing Dialogues under Dialogue Macrogame Theory
Revised August 12, 2003

The purpose of this section is to give some guidance to people who would like to analyze dialogues using DMT.

Preparing to Analyze a Dialogue:

Details of preparation will depend on the purposes of the analyst, but there are certainly some recurrent needs.

Often the amount of information needed is quite small, but having some information of each kind is generally essential.

Some division of the dialogue into units of analysis is needed. To prevent circularity of methods, it is best to identify the units before analyzing. This raises various problems for collections of dialogues; see below.

For the particular purposes of developing DMT, one of the goals is to develop a broad based collection of macrogames. Because the dynamics of dialogue varies a great deal with the situation, especially the obligations, tasks, status and expectations of the participants, a broad sampling of dialogue situations is needed. This requires moving beyond controlled laboratory settings and accessing dialogues from diverse sources.

Diversity of sources leads to a strong desire for flexibility in the analysis tools and conventions. There is a great advantage in analyzing dialogues in some form easily derived from available forms.

On the other hand, for the purpose of sharing of data and results, uniform, widely used representations are desirable. For DMT this tension is still unresolved, and the conditions of resolving it change continually. New annotation standards, new tools, new sources of data, particular comparative work (such as the Maptask - DMT comparison in the companion paper) change priorities and increase the need for flexibility.

DMT, and the collection of analyses slated for this website, have no uniform dialogue representation. Variation shows up in the treatment of timing information, links between transcript and speech tracks, and the representation of overlapped speech. For our own purposes, the turn boundaries given by the data sources have turned out to be quite usable, despite their lack of uniformity.

Compared to other kinds of linguistic study, e.g. studies of sentences, the individual items for examination in dialogue studies are huge, initial acquisition is expensive and Internet sharing, of both data and results, seems imperative.

Thus it is reasonable at the beginning of any new effort to consider how results will be shared.

It is also worthwhile to consider what tools will be used. Currently the key tool is Excel. It allows flexible manipulation of the texts, and graphic annotation either as a primary annotation method or as a reexpression of symbolic annotation.

The attached paper was prepared with analyses and diagramming in Excel 2000 v.9.0 spreadsheets, large texts in Word, some distribution files in pdf, along with an array of website tools. Very recently it was discovered that it is much more efficient to analyze dialogues by using the graphic form as soon as possible, and then later (if needed) derive the remainder of the symbolic form from the diagram.

Analyzing a Dialogue

When someone analyzes a dialogue in DMT, they are creating and affirming a collection of statements about the dialogue. They use personal judgment, knowledge of society, culture, language and individual experience to come to conclusions. The statements say, in effect, that the definitions of DMT entities fit the dialogue in a particular way. It is thus very important to have a definite, clear idea of what it means for a definition to fit a particular portion of the dialogue.

It is not self evident how to use the definitions that are given in the tables of entities. They need to be interpreted. This section gives some guidelines for that process.

Since the whole record, including audio, video and transcript, is the object being analyzed, all parts should be included when the analysis is given to others. A DMT analysis is recorded as annotation on the transcript.

Before analysis, the dialogue is usually divided into dialogue turns, although finer division can be used if desired. A turn is an interval of expression by a single participant. So the units that are taken as undivided wholes are either turns or parts of turns, such as sentences, intonation contours or auditory intervals punctuated by changes of gaze.

Notice that this definition does not enforce any alternation of turns. It merely notes discontinuities of expression on the part of each individual. (The definition fails if both individuals talk continuously. This happens, but it is rare, and transcripts of it are even more rare.)

The subdivision of the dialogue is often strongly influenced by the subdivision imposed when the recording and transcription of the dialogue were done. For a wide ranging study (such as the development of DMT), use of a diverse collection of other people's data is a practical necessity. Thus it is seldom possible to achieve a homogeniety of division of dialogues into parts. In DMT, there has been some sensitivity to this fact in creating the definitions. They allow the analyst to say "This happened here in this unit" or "That happend there in that sequence of units" without any vital use of unit boundaries. At least that has be the design intention for the definitions.

Since a turn may perform several actions, the corresponding scopes of games (contiguous sequences of turns) and unilaterals often overlap. Division into turns, which may or may not overlap other turns in part or entirely, should be done before analysis to avoid circularity. However, if during analysis, one interval of one speaker's expression is subdivided into a sequence of several turns that clearly have differing DMT status, it does not seem to incur much of such risk.

Analyzing dialogue from diverse sources inevitably requires crossing project boundaries and using data beyond project control. Analysis of dialogue from diverse sources is possible only if the amount of attention given to each source is made manageable. One practical approach is to accept the turn divisions which are provided by the dialogue source.

The activity of analysis is opportunistic. It records a plausible view, but not how that view was developed. It is therefore convenient to first mark clear cases, with their types, for example marking Media Management and Acknowledgment.

Analysis can be applied to dialogues in any language or combination of languages. The dialogue games that have been identified so far have been developed using dialogues in English. For languages other than English, and for previously unexamined situations in any language, new dialogue games may need to be defined. Since identification of use of DMT elements is nominally independent of the language used, definitions should not refer to particular words or language structures.

Scopes of evidence for application of the definitions

The advice above does not identify what portion of the transcript should be considered in identifying particular events. It turns out that this identification embodies significantly different alternatives, with serious consequences for theory. In particular, there are at least these alternatives:

  1. Whole interaction: Evidence is drawn from the whole interaction to assess the applicability of each definition at each boundary.

  2. Left to Right: Notice that this section has some unexpected features, and that DMT analysis is not strictly analysis of completed utterances.

    Each definition applies at a particular boundary, or not, based on its fit to the context available to the left, toward the past, all of the influences available to the participants at the moment represented by the boundary. This assumes that the DMT status of each utterance is determined by the form of that utterance, together with its left context.

    This approach is in contrast to the Moment of Use approach below. The one below allows interpretation of some items to be deferred, and it allows the hearer to in effect determine the act category of certain utterances of the speaker.

  3. Moment of Use: This one is less evident, and so it needs a constructed example. Suppose B is employed, and that A is B's supervisor. B is hoping to get A to agree to some extra time off the job.

    • A: How's your project going?

    • B: Very well. We're a couple of weeks ahead of schedule.

    • A: Good.

    • B: [0] My family is urging me to take a long trip at vacation time.

    • A: [X] My family talks a lot that way too.


    • A: [Y] That would depend on some details.

    The idea in the example is that A can interpret [0] as a bid of a Permission Seeking game, or not. In [X] A saw [0] as a Unilateral, and so the Permission Seeking game has not been bid. In [Y] A saw [0] as bidding Permission Seeking, and A is beginning to discuss the conditions of giving permission. [Y] accepts the bid. Both the existence of the bid and its acceptance are evidenced by A's talk about dependency, since what depends on the details is permission to take a long trip.

    If A says [X], B might try to rescue the discussion as follows:

    • B: [1] I'm glad you're sympathetic. I think I need an extra week off.

    The idea here is that in [1] B is taking [X] as an acceptance of a bid made or begun in [0]. [1] then further specifies the bid, saying more about what permission is being requested. Like the previous case, A is the speaker of [X] but B is interpreting [X] as an acceptance rather than a Unilateral.

This sort of case allows A to decide whether B bid a game, or alternatively did a Unilateral TELL. (B may not be in a social position to dispute A's decision.) A chooses the game status of [0], but B is the speaker of [0].

This kind of example suggests that in such cases the form of particular utterances in dialogue is not what determines their game status, and thus the further course of the dialogue. Rather, the determination depends on whether the response to such utterances pursues what would be the goal of the bid if it is a bid.

The DMT analyst needs to look at not only each particular utterance and its prior context, but also the immediately succeeding utterances by the other party, how it was interpreted up to the point when the speaker begins to speak again.

Beginning to speak again, the speaker must use this knowledge, which is why this option is labeled Moment of Use.

For the analyst, classification of what the speaker says, e.g. a game bid in the examples here, comes not at the end of what the speaker says, but rather at the end of the responder's showing how it was taken. At first this may clash with an analyst's habitual views of text and meaning, but it reflects the dynamics of some interactions. (Others are so clear that R has no freedom to take certain items in more than one way.)

(The case illustrated is probably the dominant case, but other things can happen.)

Notice that if a certain part of the interaction is to be labeled a Unilateral rather than a bid (or a beginning of a bid), it must make sense as some particular kind of Unilateral. If it does not, then the most plausible analysis of that part, along with the response to it, may be as a bid followed by a refusal. This makes it fairly inevitable that for certain attempts to bid a game, the identity of the game will not be discernable.

If it is clear that the participants' views of the dynamics of the dialogue later changes, beyond the view that prevails in first responses, the analysis should reflect the shared consensus hindsight view of the participants, in the analyst's confident best judgment.

Of the three options (Whole Interaction, Left-to-Right and Moment of Use), the first one is worst since it has the analyst using information that, for the participants, represents the future. The governing principle is that the analyst's judgments should be qualitatively similar to the participants' judgments in the situation. The example indicates a basis for preferring Moment of Use to the other one: Left to Right with interpretation only from the end of the text so far to the left.

So, Moment of Use is the approach used in DMT.

Some unusual structures

There is an unusual structure of double bidding that goes beyond the normal pattern. For example, someone may say

"Do you want some ice cream? What kind would you like?"

This can be analyzed as a single, somewhat elaborate bid. The question "Do you want some ice cream?" is in effect preparatory for the bid which is effectively in use, about which kind of ice cream is preferred. It loosely resembles many other sorts of preparatory uses of language, simply presupposing that ice cream is wanted. In that function, it can be related to other linguistic presuppositions.

There is another sort of preparatory action that also needs to be analyzed. Someone may say

"May I ask a question?"


"May I say something?".

Neither of these is specific enough to identify a particular game being bid. Either one could be rejected, as in

"I have to leave now."

If there is rejection, they can be treated as rejection of a bid of an unknown game. If there is acceptance, the game can be identified, and the sequence treated as a two step bid action. This approach is preferable to analyzing the sequences as using embedding.

Goals in DMT definitions

As noted in the section on Macrogames, a macrogame is defined by specifying three goals and the name of the game. The names are convenient phrases for refering to the games. The names should not enter into the judgmental process of deciding how or whether a particular definition fits. Even after making this name independence clear, experience has shown that it is hard to create a set of names that is neutral and never misleading. So users of DMT are warned to be very cautious in relying on the names of any of the entities.

Several points need to be made about the interpretation of each goal:

Many of these understandings apply to Unilaterals as well as Macrogames.

Goal Manipulation Actions

Dialogue games are in essence groups of goals that are evoked, activated and deactivated together. As noted in the section on game acts , DMT has six basic acts which participants can use to change the state of activation of games. There is bidding a game, abbreviated bg, and there are ag, rg, bt, at, and rt.

Bidding, rejecting and accepting are often entirely implicit, identifiable based on the other activities in dialogue. Discerning these acts often depends on understanding the social situation in which they occur. For example, in tutoring, certain games are reserved for the teacher. The student lacks the right to initiate them.

A game can last indefinitely, i.e. the goals of a game can be active indefinitely, provided that the joint goal of a game continues to be pursued. However, in practice, games are often very short. (The possibility of long games distinguishes this notion of dialogue game from certain other technical constructs that have similar names.)

Goals and Their Status

Analysis depends especially on identifying the goals that are brought into the dialogue, both implicitly and explicitly, and on judging their kind and status at each point. We will discuss goals in two categories: individual goals and joint goals.

We will talk about the status of goals relative to each dialogue participant, and also about a sequence of acts that can produce that particular goal status (sequences in parentheses). For a particular participant and goal, a goal is:

1.        not evidenced,
2.        proposed, (bg)
3.        rejected, (bg, rg)
4.        active, (bg, ag)
5.        satisfied, (bg, ag, bt)
6.        exhausted (unsatisfied and unpursuable), ( also bg, ag, bt)
7.        dismissed. (bg, ag, bt, at)

Except for being satisfied or exhausted, these conditions exclude each other. A goal G is active while it is being pursued, and also while it has one or more subgoals that are being pursued as a way of pursuing G.

Missing Games and Overlapping Games

Analysis sometimes encounters a point in a transcript where some particular sort of act, such as bidding a game, is clearly done, and yet no game on the list fits. Such events are to be expected on several grounds:

At such points it is appropriate to

  1. define a new game
  2. enlarge the scope of an existing game

Similarly, there are situations where choosing a particular game among alternatives is difficult. Such situations are unavoidable even in principle. Some require resolution, and some do not. Notice that this is a problem for the analyst but not for the dialogue participants. The participants may be able to interact flawlessly without thinking about the analyst's alternatives.

At such points in analysis it is appropriate to

  1. make an arbitrary choice, seeing the alternatives as equally applicable and not defective
  2. introduce a distinction into one or more games which reduces the potential for analyses to find the particular alternation

Notice that in general it is the system, the collection of definitions, that must be improved, not simply individual games.

Similar considerations apply to finding Unilaterals in analysis.

Incoherent Dialogues and Motivational Coherence

It is common to find dialogues which show interaction but no collaboration. An exchange of Unilaterals can take place, with some of them prompted by the previous unilateral, but no joint goals are identifiable. This is not necessarily considered to be incoherent.

However, if there is a game open, and thus there are joint and personal intentions being pursued, then the games that arise are expected to cohere with the games already open. More precisely, the joint intention of each embedded game is expected to advance the joint intention of the game use which immediately contains it.

During analysis, if in your judgment this seems to be what actually happened, then that should be pointed out as part of the analysis. Otherwise, showing an embedded game, a game pursued within the scope of another game, the inner game is assumed to serve the next outward one. This subordination of goals, in DMT but also in other examinations of goal pursuit activity, is called Motivational Coherence.

Media Management:

At the beginning of a dialogue, participants are not communicating with each other and there are no active goals. Using language along with non-linguistic means, one (or occasionally both) initiates the interaction. Such initiations of interaction are not accidental; they are intentional, i.e. they are actions performed in pursuit of goals, necessarily a goal to interact, and generally other personal goals that interaction may help to achieve. Until a joint goal can be proposed and activated, all of the active goals are individual goals, seeking to initiate interaction.

There are varieties of language that are nearly always confined to this task of initiating interaction. Greetings, shouting of names, self identification in telephone calls and other comparable actions are dubbed Media Management.

Other sorts of actions are used to end an interaction. For example, there is a recurrent pattern of six actions that is found at the end of many telephone conversations. It arises when there are no active pursuable joint goals. Each element of the pattern is a unilateral. The pattern as a whole has no status in DMT.

Whether or not the conventional forms are used, the lack of joint goals and the distinctive activity of beginning or termination of interaction distinguishes Media Management from dialogue game activity. Despite appearances above, often there is no sharp boundary between media management and dialogue game activity. For example, elaborate self identification may include as well the specification of joint goals for pursuit.


In immediate interactions, (but not in exchanged letters and email), a very wide range of actions can be used by participants to acknowledge actions by other participants. Forms such as OK, Right, Um Hmm and Yup are common, but the list of possible forms is unbounded. Echo-like forms are also common. Acknowledgment may express agreement, or understanding, or mere reception. Often there is no way to tell which of these a particular instance represents. Distinctively, acknowledgments do not work to introduce goals, nor do they remove goals from being active.

It is often convenient to treat the entire dialogue as a use of the Conversation Seeking Game, bid by the one who uses the first words. This makes little difference except in cases where one party wants to terminate the dialogue and the other does not. In such cases, it is convenient to identify certain actions as bidding the Conversation Seeking Game and rejecting that bid.

A common exception arises when an excerpt is analyzed. There the establishment or termination of the Conversation Seeking Game are missing, and so there is not an adequate basis for assuming it. A comparable case arises for interactions that use continuously open media.

Identifying Unilaterals: This is in many ways comparable to identifying bids of games. The surrounding text needs to be taken into account, including the response of the recipient. If the recipient did not treat a statement as part of a collaborative interaction, this is a strong reason to treat it as a unilateral.

It is very clear that reliable unilateral identification cannot always depend on linguistic form. However, more than for bids of games, conventional forms provide strong evidence for certain kinds.

Back to the Definitions of Particular Games page: [ Definitions of Particular Games: ]

Back to the Abstract Definition of Games page: [ What's a Game? ]

Back to the Entities and Terminology page: [ words and things ]

Send email to Bill Mann: bill_mann@sil.org