Using Speech Act Theory

Title: Using Speech Act Theory to Model Conversations for Automated Classification and Retrieval

APA Reference

Twitchell, D. P., Adkins, M., Nunamaker, J. F., Jr., & Burgoon, J. K. (2004, June 2-3). Using Speech Act Theory to model conversations for automated classification and retrieval. Proceedings of the International Working Conference Language Action Perspective Communication Modelling (LAP 2004), New Brunswick, NJ, pp. 121-130.

Abstract

Instant messaging, chat rooms and other forms of synchronous computer-mediated communication (CMC) are increasing in use in the business, military, and consumer world. The language action perspective provides methods for analyzing and modelling repeated business conversations including synchronous CMC. This paper describes a method for creating a profiles for large amounts of synchronous CMC conversation after it has occurred. Called a speech act profile, it is based on speech act theory. The profiles can be used either as patterns for classifying conversations or for creating visual maps of the conversations themselves. Application of the profiles in information retrieval and deception detection are discussed.

Keywords

see terminology below

Authors' Bio (name, school)

Douglas P. Twitchell, ISU
Mark Adkins, at industry
Jay F. Nunamaker Jr., CMI
Judee K. Burgoon, CMI

Problem Statements/Phenomena

"[The] ever-growing archive of conversations represents an opportunity for both language action researchers and business managers. Researchers can use this data to study human communication in chat/IM conditions. Managers can use the data to detect fraud, retrieve conversations of interest, and direct training." pg.2

This paper describes the theoretical foundation to speech act profiling and potential uses for the tool.

Research Questions

No explicit research questions. The paper is a descriptive paper of Speech Act profiling and the theoretical models it is based on with a brief explanation of the foundational theories leading to Speech act profiling.

Theory Used or Developed

Language Action Perspective (LAP), which models business conversations, is based on two theories: Speech Act theory and GoldKuhl's predefines speech act patterns, which are not described in this paper.

Hypothesis, Independent Variables, Dependent Variables

These following claims are suggested but not tested in this paper.

Because deceivers are loath to be caught in their deception, they will often put on a submissive front. The expression of fewer assertions and more expressives (which include agreements) could indicate submissiveness. pg.3

Furthermore, because deceivers usually express some uncertainty as a way to hedge their deception, those looking for deception could issue a query for conversations where a participant was being submissive or showing uncertainty. pg.3

Terminology

language action perspective (LAP)
modeling of business conversation
Speech Act Theory
asserts that with each utterance in a conversation an action is performed by the speaker
speech act profile
visualization tool to see results of speech act theory
SAMPO, DEMO, BAT, The Coordinator, MILANO, Negoisst
alternative LAP software to speech act profiler
perlocutionary act
(of a speech act) producing an effect upon the listener, as in persuading, frightening, amusing, or causing the listener to act. deception is an example. look it up
illocutionary act potential
model of dialog by Alston (2000). pertaining to a linguistic act performed by a speaker in producing an utterance, as suggesting, warning, promising, or requesting. look illocutionary up
dialog act modeling
method to visualize conversations by Stolcke et al. (2000)

Methodology

Method Type: descriptive argumentation

Description:

The paper defines how speech act profiling is done.

These methods [dialog act modeling and illocutionary act potential] create a set of summed probabilities for each speech act type during a conversation, which are then subtracted from the training corpus average to obtain divergences from normal speech. The speech acts used are the 42 dialogue acts in the modified SWBD-DAMSL tag set (Jurafsky, Shriberg, & Biasca, 1997). With the speech act potential probabilities and the categories of speech acts defined, a visual representation of the speech act profile can be created [in a radar graph].

speechactradar.png

Subject and Selection Criteria: not applicable

Sample Size: not applicable

Measuring Instrument:

set of summed probabilities
unknown

Major Findings

not applicable

Discussion Summary & Author Recommendations

Authors define the benefits of speech act profiling as the following:

"Instead of predefining conversation types and attempting to design or facilitate business conversations according to the categories as done by traditional LAP systems, speech act profiling looks at business conversations as existing empirical data that are categorized during or after they actually take place." pg. 2

Recent research using speech act profiling was able to identify uncertainty in chat participants who were instructed to be deceptive (Twitchell, Nunamaker, & Burgoon, 2004). Speech act profiling’s ability to classify participants as uncertain may aid those searching for deceptive behavior. pg5

Other uses of speech act profiling are the following:

  1. Attach intent profiles to conversation
  2. combining symantic meaning with search engine querries
  3. classify conversations into a taxonomy; Winograd (1987) LAP classifications are as follows with identify purpose or intent:
    1. conversation for action
    2. conversation for clarification
    3. conversation for possibilities
    4. conversation for orientation

Hidden Markov model : alternative classification model of communication, http://www.csse.monash.edu.au/~lloyd/tildeMML/Structured/HMM.html

Speech act profiling is one method that has been shown to be useful in discriminating between types of conversations including conversations with deception- induced uncertainty. It, along with other similar Markov-model-based techniques, may be useful in classifying conversations into more general framework, giving researchers and managers the tools to manage the millions of conversations produced each day. For researchers, not having to code conversations for intent by hand is an important labor-saving feature of conversation classification.

Why paper is important? Why paper is cited? '

  1. describes the history and theoretical foundation of speech act profiling
  2. lists numerous foundational researchers
  3. Anyone continuing the research using speech acts profiling should be very familiar with all the theories and authors mentioned in this paper.

Persistent Link to Library

not available, see the AFOSR repository at CMI

Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-Share Alike 2.5 License.