AUTHOR: Sassen, Claudia
TITLE: Linguistic Dimensions of Crisis Talk
SUBTITLE: Formalising structures in a controlled language
SERIES: Pragmatics and Beyond New Series, Volume 136
PUBLISHER: John Benjamins Publishing
YEAR: 2005
Sally Wellenbrock Hinrich, Department of English, Oklahoma State
University
DESCRIPTION
Using a computer-based modeling system, the author proposes Head-
driven Phrase Structure Grammar (HPSG) -based discourse grammar
for a restricted language that allows the identification of well-formed
discourse patterns. The modeling system is designed to analyze and
identify discourse used to coordinate actions that might result in
avoiding a potential disaster. A corpus for analysis was drawn from
aviation accident transcripts between cockpit crews and between flight
crews and controllers during an emerging crisis situation.
SYNOPSIS:
Chapter 1: Towards an analysis of crisis talk.
Using the domain of aviation communication, the author defines Crisis
Talk as ''a dialogue genre that occurs in threatening situations of
unpredictable outcome, with no obvious way out, and requiring
spontaneous decision, unconventional strategies and unrehearsed
actions.'' Sassen's primary objective is to establish a framework that
examines the viability of incorporating a speech-act methodology in
order to categorize the linguistic structures occurring in aviation
disasters. The corpus used for analysis is drawn from air traffic control
and cockpit voice recorder transcripts available on the internet from
independent sites.
The analysis focuses on the functions found in the linguistic
sequences of the interactions rather than on individual lexical items
within utterances. In order to accomplish the analysis, extensible
markup language (XML) is combined with an extended form of HPSG-
based attribute value matrices. Through this extension of the formal
HPSG annotation system, Sassen argues that both utterance-level
and discourse-level communication can be modeled using a single
HPSG-based sign.
Sassen predicts the following results from the analysis:
1. Crisis talk is different from non-crisis talk with respect to
interactional patterns;
2. In order to disambiguate speech acts, the model of Searle &
Vanderveken's illocutionary logic requires more precision with regard
to the propositional content that is presupposed. This extension can
be achieved by an HPSG-based formalism;
3. An extended HPSG-formalism is an adequate model for the
representation, description and explanation of the disambiguation of
illocutionary acts;
4. It is possible to extend HPSG-based structures and principles from
the interactive sentence level to the utterance level. An extension of
the HPSG-based inventory from the utterance level to the discourse
level is also possible: speech act sequences may be modeled by one
HPSG-based sign;
5. XML uses an attribute value archiving and retrieval formalism, and
is potentially flexible enough to fulfill the requirements of crisis talk
annotation.
The final section of Chapter 1 describes relevant regulations for
language used in aviation, more commonly called ''air traffic
communication.'' As part of the background on sources of error in
communication, Sassen notes that ''among the factors unequivocally
attributable to language, ambiguity figures as a main error source'' as
a result of errors in lexical choice and in grammatical structure. It is
this ambiguity that is targeted by the methodology discussed in the
following chapters. The final section in this chapter describes the
function of Cockpit Voice Recorders (CVRs) for analysis of aviation-
related accidents. The transcripts created from recovered CVRs are
the basis of the data to be used in this study.
Chapter 2: Discourse-related approaches
In explaining the rationale of choosing illocutionary logic as the
foundation for the study, Sassen argues that illocutionary logic is an
adaptable, adjustable theory and the parts of the theory which do not
apply need to be discarded. Building on this argument, a discussion
showing Austin's (1962) work on illocutionary verbs, is followed by a
section describing Searle's elaboration of simple vs complex
illocutionary acts, semantic rules, and input-output conditions as ''a
pre-requisite for every kind of speech act which pertains to 'intelligible
speaking' and 'understanding''' (Searle, 1969; Searle & Vanderveken,
1985). Because Searle & Vanderveken offer no syntactic system to
identify illocutionary force indicating devices (IFIDs), nor do they
describe a suitable grammar, Sassen states that their illocutionary
logic system ''must be extended'' in order to adequately frame a theory
of discourse. Additional theories of discourse are outlined that could
serve to extend the illocutionary logic framework, including the
advantages and disadvantages of incorporating features from
conversational and discourse analysis.
Chapter 3: Linguistic and corpus methodology
Opening with a discussion of inductive and deductive methods,
Sassen observes that conversational analysis, as an inductive
method, ''has to be precluded'' because the analytical framework must
be established before applying that framework to empirical data.
Sassen then examines a methodology to formalize Searle's theory in
order ''to make his results and extensions to his research efficiently
processible.''
The analysis of the aviation accident transcripts draws from the
domain of linguistics incorporating two theories: speech act theory
which relates to illocutionary components of utterances that is
combined with syntactic theory which addresses sentence-level
components. Both theories are then joined through the illocutionary
force indicating devices. Reiterating her goal of ''providing
foundational research for the remote goal to (semi-) automatically
identify speech acts through an annotation system that identifies
specific meanings to a sequence of linguistic structures,'' she then
proposes the validity of transforming grammatical units into attribute-
value structures. This transformation is necessary because it allows
for a wide range of potential structures while it also represents the
pragmatic features of sentences. Sassen discusses how this
transformation is accomplished through head-driven phrase structure
grammar (HPSG) which is ''enriched by certain attributes that are not
as yet common to HPSG grammars,'' but which will provide a means to
examine how well the approach yields significant results. Sassen then
explains how HPSG-formalism allows a detailed description of the
syntactic structure of sentences and also their semantic treatment,
including various pragmatic and background features.
In the next section describing the creation of the research corpus,
(composed of two of the 77 aviation accident transcript files), Sassen
outlines the transcription conversions necessary to accommodate
analysis of the raw transcript data. In addition, ''[a] crisis talk markup
system clearly needs...detailed syntactic, semantic and pragmatic
features'' which can be more accurately described by an extensive
markup language (XML), which allows data to be structured into two
fields: ''object language/concrete data'' and ''abstract units/meta
language.'' XML was chosen primarily because it allows both
hierarchical and sequential ordering of elements. Several approaches
to annotating the linguistic structures are presented, including
diagrams, based on a system developed by Bird & Liberman (1999).
Chapter 4: Analysis of general dialogue properties
The criteria for marking the transcripts are based on guidelines
developed by the Expert Advisory Group on Language Engineering
Standards (EAGLES; Gibbons, Mertins & Moore, 2000). Examples of
this process of converting the ATC/CVR transcripts to the transcription
criteria are presented in combination with references to the EAGLES
guidelines. One advantage to this annotation system is the ability to
identify and document overlapping transmissions as an ongoing
relationship to the developing dialogue (rather than as sequentially
listed lines of dialogue/events typically used in the aviation
transcripts). Most of the aviation accident transcripts used sentence-
level, or complete word forms in transcription, which ''...has the
advantage that annotation and retrieval tools may be applied relatively
unproblematically to speech as well as to writing.'' Sassen also notes
that the EAGLES guidelines support using standard (''dictionary'')
spelling ''...[which] has the advantage of improving readability for the
human user and of increasing processibility for taggers and parsers.''
Following the conversion of the transcripts, perl-scripts are used to
generate data as a text base and XML-markup. Sassen notes that no
typical development of phases that lead to a crisis can be observed in
the transcripts, so she assigned three broad phases and supports her
choices with examples from the research data: ''non-crisis,'' ''before
the crisis'' and ''crisis.'' These broad phases of aviation communication
each include subdivisions of ''conversational phases'' (opening, medial
and terminal). A separate categorization includes discourse-control
processes that consist of three potentially overlapping functions: topic
processes (goal-oriented sequences of information exchange), uptake
processes (error-control strategies in support of the topics) and
framing processes (orientation points within the discourse). Discourse-
control processes are then examined using examples of both
professional (related to the aviation situation) and non-professional
(not related to aviation situation) communication.
Chapter 5: Analysis of particular dialogue properties
In the introduction to this chapter, Sassen states that ''For an analysis
of crisis talk, not every ATC/CFR-transcript is appropriate. To fulfill the
criterion of empirical soundness, the transcript is required to show
both threads of crisis talk and threads that do not apply to a crisis
situation... [these threads are] essential for contrastive analyses of
crisis and non-crisis talk features. The other essential requirement is
the criterion of empirical completeness of the dialogue...the present
analysis stresses the development of a method that allows
investigation of crisis talk from a *functional* point of view, i.e. the
analysis of language use....there is a need for syntactic analysis, too,
as it is a foundation of an automatic analysis.'' A discussion of the
types of regularities in the two transcripts includes the decomposition
of both the dialogue and separate speech acts. Problems in
automating the transcripts are explained, including coding choices
made to represent speech sounds as well as other non-vocal sounds
and for minimal sequencing (such as command-response) and how
certain omissions in the original transcripts were coded. There is
discussion of the system of annotation and graphic representation of
the transcript speech as well as a general description of how
sequencing of voice and sounds could be noted.
The following section outlines the criteria for choosing the two
transcripts that were analyzed with the system. This section describes
how an utterance sequence (from a transcript) is assigned the HPSG-
based notations, how the representations were achieved; then moves
on to the implementation of XML as a denotational semantics for
HPSG-based signs. In addition to a careful explanation of the coding
system, a graphic model of a generic HPSG-based sign
representation model and an example of HPSG-based structure for
the phrase ''disconnect the autopilot'' demonstrate the multiple levels
of annotation. The analysis section ends with a brief explanation of
how HPSG-based signs are mapped into XML.
Conclusion (included in Chapter 5)
An extended form of HPSG provides the means to account for
linguistic principles and rules that determine the well-formedness of
linguistic expressions while XML provides the framework to
incorporate the aviation transcripts which are ''real-life'' situations.
Sassen observes that ''Ideally, this analysis can be used to minimize
escalations during flights and to make aviation safer.'' This section
ends with suggestions for refining a theory of discourse structure
which goes beyond modeling a controlled language such as aviation
discourse.
The overall quality of the publication is excellent. The appendices
include: glossary of aviation terms; abbreviations used for the matrix
of HPSG-based sign; atomic (abbreviated) representation of speech
act types; minimal sequences (of the coding system) and their
modifications; two sample transcripts; and background information
related to the transcripts. There is a useful subject index and an
extensive reference list. A bit of confusion may occur when consulting
the reference list as two pages of references are out of alphabetic
sequence.
CRITICAL ANALYSIS
Several times, Sassen alludes to the problems which prevented her
access to the original accident audio recordings for a more complete
analysis. However, she provides well-documented information related
to the transcribed dialogues for the sources she uses in the analysis.
Due to increased security measures in the United States, access to
several sources of aviation discourse has become more challenging,
often requiring lengthy approval procedures.
Sassen's linguistic theory draws from both European and American
sources. For scholars interested in developing a richer background in
discourse-related analysis from several points of view, there may be
some non-aviation-related references which could prove useful in
future linguistic research.
Once past the original assignment of speech acts to the lines of
dialogue, it appears there is still an enormous amount of work that
must be done to make the transcript data useful for machine-based
analysis. Sassen argues that this analytic system ultimately
provides ''knowledge of possible defects'' of language used in crisis
situations. This approach is relies on restricting aviation
communication to the rules of aviation phraseology. Yet, as many of
the discarded transcripts show, much of the ''crisis'' discourse is, in
fact, not particularly well-formed, but it also is not
necessarily ''defective'' just because it is not grammatically accurate. In
daily communications between tower and plane (available via the
internet), there are frequent instances where pilots and air traffic
controllers do not use just the approved phraseology, yet the planes
arrive safely at their destination.
Since cockpit voice recorders and flight data recorders operate a
continuous tape loop lasting for only 30 minutes, the final minutes of a
tape may contain only non-linguistic sounds (such as mechanical
failure, coughing related to heavy smoke, etc.), thus minimizing the
amount of discourse available for this type of analysis. In order for any
transcripts to be accepted into the corpus, Sassen argues that the
transcript dialogue must have ''empirical completeness.'' Readers with
conversational or discourse analysis backgrounds may question the
requirement that all utterances must be ''well-formed'' or ''complete'' in
order to provide valid linguistic evidence of a developing or ongoing
crisis. These ''incomplete'' dialogues may, to some degree, contain
verbal information that suggests the participant(s) not only recognize
but are attending to emerging problems or impending danger (Goguen
& Linde, 1983).
While ambiguity may be a common source of language error, as
discussed in Chapter 1, Sassen does not address the means by which
American and international formal aviation phraseology attends to and
minimizes these ambiguities (Sanne, 1999). Further, only two of 77
transcripts (available on the web) met the requirements for the
analytic system used by the author. There was no mention as to how
the author would propose to evaluate the remaining 75 transcripts. It
seems problematic that this many transcripts of authentic ''crisis''
language should be discarded if the research objective is to improve
aviation safety by looking at the discourse of accidents.
In conclusion, Sassen's methodology may prove valuable to
researchers working in machine-based linguistic analysis, whether or
not they are involved in examining a ''controlled language.'' Sassen's
innovative methodology and research adds to an increasing body of
research that investigates the discourse properties of the language of
aviation while it addresses the critical issue of improving safety in
aviation.
Note: For those unfamiliar with HPSG, I recommend reviewing this
website prior to reading chapters 3-5:
http://en.wikipedia.org/wiki/HPSG.
REFERENCES
Austin, J. L. (1962). How to do things with words. Oxford: Clarendon
Press.
Bird, S., & Liberman, M. (1999). A formal framework for linguistic
annotation. (Technical Report). Department of Computer and
Information Science, University of Pennsylvania (ms-cis-99-01).
Gibbon, D., Mertins, L., & Moore, R. (2000). Handbook of audiovisual,
multimodal and spoken dialogue and systems resources and
terminology for development and product evauation. Doordrecht, New
York: Kluwer. (Final Report of LE EAGLES Phase II project (LE3-4244
10484/0) for the European Commission).
Goguen, J., & Linde, C. (1983). Linguistic methodology for the
analysis of aviation accidents. (No. NASA Contractor Report 3741.
[NTIS No. N84-15135]). Moffett Field, CA: NASA Ames Research
Center.
Sanne, J. (1999). Creating safety in air traffic control. Linkoping,
Sweden: Arkiv Forlag.
Searle, J. (1969). Speech acts. Cambridge University Press.
Searle, J., & Vanderveken, D. (1985). Foundations in illocutionary
logic. Cambridge University Press.
|
