Linguistic Factors in Cockpit Communication

1
Linguistic Factorsin Cockpit Communication

• Manfred Krifka
• Humboldt Universität, Berlin
• collaborating with
• Silka MartensFlorian SchwarzCarrie Clarady

2
Research on Cockpit CommunicationSome Background

• Social dynamics of group interaction as an
  important factor in the origin of accidents
  and in the success of dealing with accident-prone
  situations (cf. Robert Helmreich, Managing
  Human Error in Communication, Scientific
  American 1997.)
• Crew Resource Management (CRM) Rating and
  improving of crew performance in aviation and
  other fields in which professional groups
  interact in situations with high task load.
• But over and above the general social dynamics
  of group interaction, there are specific
  problems relating to language and
  communication in such settings.

3
Cockpit Communication

• Possible areas of research
• Problems of language (e.g., the structure of
  human language, features of the technical
  language).
• Problems of communication (that is, problems
  of language use)

4
Cockpit Communication Problems of Language

• Cf. Stephen Cushing (1994), Communication
  Clashes and Aircraft Crashes.
• Misunderstanding of We are now at takeoff
  implied in Tenerife accident in 1977
• Problems of sentence parsing back on the
  power back on the power
• Problems of phonological identification climb
  to five zero climb two five zero

5
Cockpit Communication Problems of Communication

• Mitigation of contributions especially by the
  lower-ranked pilot (First Officer or Engineer)
  for face-serving purposes.
• Not very much more fuel. (Engineer to Captain
  five minutes before engine stopped, United
  Airlines, Portland, 1978)
• (cf. Charlotte Linde, The quantitative study
  of communicative success Politeness and
  accidents in aviation discourse, Language in
  Society 17)

6
Cockpit Communication Research Techniques

• Observation of real situations
• Transcripts of communication before and during
  accidents (recorded by Black Box and
  published by National Transportation Safety
  Board)
• Observation and ratings of communication of
  regular flights.
• Important data, but difficult to come up with
  generalization from particular cases.
• Observation of communication in flight
  simulators
• Experimental scenarios.
• Similar scenarios for different crews.
• Advanced simulator techniques guarantee
  closeness to reality.
• But Simulator sessions extremely costly.

7
Group Interaction in High Risk EnvironmentsLingu
istic Factors
GIHRE Subproject, sponsored by Gottlieb Daimler
and Karl Benz Foundation. First project phase
at Department of Linguistics, University of
Texas at Austin, with Carrie Clarady.
8
Linguistic Factors First Project Phase
Basic Questions Correlation Task Load ?
Communicative Behavior Correlation Crew
Performance ? Communicative Behavior
9
Linguistic Factors First Project Phase

• Data
• Simulation flights on B727 with similar
  scenario, carried out by NASA, 1987
• Transcribed by University of Texas Aerospace
  Crew Research Project(thanks to Bob Helmreich
  and Brian Sexton)
• Every flight has 4 transcribed segments A
  climbout from SF, descent and landing into SAC
  Medium task load.
• B Descent, approach and landing at LAX,
  with runaway trim, jammed stabilizer, low oil
  pressure High task load. C Descent to SAC,
  missed approach Medium task load. D
  Diversion to SJC, with hydraulic malfunction,
  diversion to SFC, split flap malfunctio High
  task load.
• Performance of crews and crew members was rated
• We investigated 5 ( 1/2) flights (2
  well-performing crews, 3 poorly performing
  crews) (alltogether 9 hours, 6900 thought
  units)

10
Communication DensityThought Units per minute,
Crew 4
Segment A Segment B Segment C
Segment D
No difference in speech time well-performing /
poorly performing crews
11
Factors
Presentation of Data Segments A Medium (task
load) B C D High (task
load) Crews 3 Poor (crews) 4 5 8 13
Good (crews) Data are given in proportion to
thought units (chosen by transcribers)
12
Speaker and Addressee Roleper Thought Unit

• Captain assumes speaker role most often.
• Engineer assumes speaker role more often in high
  task-load segments.

13
Speaker and Addressee Roleper Thought Unit

• Engineer is adressed more often in high
  task-load segments.

14
Reference to Participants

• Slightly more reference to group in good crews,
  high task loadscf. Sexton, J.B. Helmreich,
  R.L. (2000). Analyzing cockpit communication. The
  links between language, performance, error, and
  workload. Human Performance in Extreme
  Environments 5, 63-68.

15
Speech Act Classification

• Question Correlation Task Load / Group
  Performance ltgt Speech Acts?
• Classical Speech Act Classification Austin,
  Searle.
• Representatives
• commit the speaker to truth of proposition, e.g.
  asserting, concluding
• degree of commitment can vary, e.g. saying,
  hypothesizing
• may relate to other parts of discourse, e.g.
  replying
• Directives
• speaker tries to get the addressee to do
  something, e.g. requesting, questioning
• degrees can vary, e.g. suggesting, commanding
• Commissives
• commit the speaker to an action, e.g. promising,
  threatening, offering
• degrees can vary, e.g. promising, guaranteeing
• Expressives
• express a psychological state, e.g. thanking,
  apologizing, welcoming
• Declarations
• changes the institutional state of affairs, e.g.
  baptizing, marrying, declare a person guilty

16
Problems of Speech Act Classification

• Existing classifications are difficult to apply
• Difference between form and meaning
• Direct command Close the door.(Command
  Imperative)
• Indirect command The door should be
  closed.(Command Declarative)
• Indirect command Why dont you close the door?
  (Command Question)
• Very indirect command Its cold in here. (not a
  command, but a statement (please), but may have
  similar action implications as commands)

17
Problems of Speech Act Classification 2

• Individuating speech acts
• ABC 801, that turn looks like its going to take
  you through, so continue the right turn to
  180.(Command, or Assertion Command, or
  Attention-Call Assertion Command?)
• 8.1530 Im on the air, so you get it.(Command,
  or Assertion Command, or Motivated Command?)
• Neglect of communication-regulating acts
• Call for attention, acknowledgements, repetitions
  etc.do not figure prominently in speech act
  classifications
• Some investigation of such features in
  Conversation Analysis (Reference)

18
Problems of Speech Act Classification 3

• Multi-dimensional nature of speech acts
• A question can be related to a statement,a
  command, a prior statement, etc.Whats the
  weather like in Sacramento?What should I do
  now?What did you say?
• A command can be related to an action, a
  statement, a question, a prior statement,
  etc.Now you fly the airplane.Tell me whats
  the weather like in Sacramento.Ask him what the
  weather is like in Sacramento.Tell me again,
  please.

19
Our Speech Act Classification

• A pragmatic approach What is potentially
  predictive for crew performance?
• Does not follow formal, but functional criteria,
  if the intended function is clear.Why dont we
  go to the book and see what you can do on it? is
  classified as a command, not a question.
• Classifies each previously identified thought
  unit as belonging to one speech act category.

20
Our Speech Act Classification (2)

• Combines speech act types (e.g., assertion of a
  proposition) with content features (e.g., report
  of action, report of previous report, prognosis,
  diagnosis)
• Factors out classification of question /
  answerYou got any problems classified as
  statement / questionYou want me to brief with
  you? as command / question
• Assumes a category of regulatives that are
  concerned with the proper flow of information

21
Speech Act Categories
SearlesClassificaiton Represen-tatives Direc
tives Commissives Expressives Regulatives
22
Examples of Speech Act Types

• Status Reports
• 4.17 UH ROGER UH WE ARE OUT AT 1443 AND OFF AT
  53 FROM SAN FRANCISCO
• 4.62 UH, ABC 801, BE ADVICED THE METER LOCATOR
  OUTER MARKER IS UNRELIABLE, ILL CALL YOU AT
  METER
• 8.28 WE HAVE AN UNDEREXCITED FAULT LIGHT ON,
  UH, NUMBER 3
• 8.89 OKAY, WERE COMING TO 350 DESCENDING TO 5,
  ABC 801
• Reports of Action
• 4.75 I KEEP LOOKING FOR THE ALTITUDE RIGHT HERE
  AND SEE 801
• 4.111 ILL PUT IT ON THE MISSED APPROACH
  ALTITUDE
• Reports of Report
• 4.461 WELL, HE TOLD US BE READY FOR ILS 24
  RIGHT
• (Not used for reciting from manuals, or relaying
  information in general)
• Prognoses
• 4.373 OKAY, SO I THINK IF WE DO THE TWO ENGINE,
  THE TWO GENERATOR OPERATION THATLL BE UH, THE
  NEXT THING.
• Diagnoses
• 8.111 I THINK WHAT THAT LIGHT WAS, WAS WHEN
  THAT FIELD, FILED RELAY TRIPPED.

23
Examples of Speech Act Types 2

• Commands
• 8.94 HEY, WITH THIS FAULT LIGHT, WOULD YOU
  SWITCH ESSENTIAL POWER OFF OF NUMBER 3, PLEASE
  FOR ME
• 8.212 ABC 801, THAT TURN LOOKS LIKE ITS GOING
  TO TAKE YOU THROUGH, SO CONTINUE THE RIGHT TURN
  TO 180
• 8.356 PRESS ON
• 8.64 OKAY, NOW YOU FLY THE AIRPLANE
• 8.224 WHY DON'T YOU MAKE IT 190.
• 8.305 FINAL FLAPS SHOULD BE 30
• Permissions
• 8.19 UNITED 801S CLEARED FOR TRAFFIC. YOU CAN
  DESCENT AT PILOTS DISCRETION.
• 8.15 OH, WHATEVER YOU WANT TO DO HERE, BOB
• Complies
• 8.414 HOW ABOUT THAT BRAKE? NOW TRY IT. NO.
• Report of Intention
• 8.32 MEANWHILE ILL TAKE CARE OF THE
  APPROACH/DESCENT HERE
• 8.63 WELLWELL TAKE CARE OF THAT

24
Examples of Speech Act Types 3

• Expressives
• 4.185 UH, SHIT
• 8.639 LITTLE DEVIL
• 8.904 BOY THAT IS A LOT OF PRESSURE
• 8.989 GREAT, EXCELLENT, NO, EXCELLENT
• 8.1019 COME ON, BABY, COME AROUND HERE
• Acknowledgements
• 4.13 ABC 801, EXPECT ILS 16 APPROACH TO
  SACRAMENTO OKAY.
• 8.410 IL JUST CRANK IT AND YOU TELL ME WHEN
  TO STOP. YEAH.
• Affirmations
• 8.1075 WELL WE GOT WERE COMMITTED TO THE
  RIGHT THATS RIGHT
• 8.1120 CANTT DO IT NO AGREEMENT.
• 8.1169 SHOULDNT BE THAT MUCH OF A PROBLEM.
  NOPE, NO NO NO.
• Rephrases
• 8.50 ONE SIX RIGHT, APPROACH. ONE SIX RIGHT,
  YEAH.
• 8.1640 GROUND SPOILERS, OUT SPOILERS OUT
  SPOILERS.

25
Speech Act Classification
More prognoses / diagnoses in well-performing
crews.
26
Speech Act Classification
More reports of intention in well-performing
crews.
27
Speech Act Classification

• Slightly more questions in well-performing
  crews.
• More questions are answered in well-performing
  crews (0.54 in poorly performing crews, 0.77
  in well performing crews)

28
Speech Act Classification

• Fewer commands in well-performing crews.
• Fewer explicit complies in well-performing crews.

29
Speech Act Classification

• Fewer expressives and emotional words in
  well-performing crews.
• Fewer expressives and emotinal words in
  situations of high task load for poorly
  performing crews.

30
Limitations of Study

• Correlation Segements / Task Load only very
  coarse.
• Number of crews analyzed far too low for any
  significance measure.
• Quality of transcripts rather varied, e.g. no
  information about paralinguistic features.
• Information about flight scenario and precise
  timing of malfunctions was rather limited.
• Revisions of speech act categories necessary.
• No classification of higher-order features of
  communication.

31
Examples of features that were not encoded
Multiple rephrases (Crew 8) 8.146 E-C CLIMB TO
500 FEET, PROCEED DIRECT SACRAMENTO VOR. HOLD
SOUTH 180 RADIAL, 5,000 FEET. C-E 5,000? E-C
YEAH. C-E AFTER 500, DIRECT TO VOR TO
HOLD? E-C RIGHT. C-E OKAY, GOT IT Invoking
Crew Resources8.1511 YOU CAN, YOU CAN LISTEN
WITH ME, I MIGHT MISUNDERSTAND. 8.2167 WHAT
ELSE? WHAT HAVE WE MISSED? Prioritizing goals
(Crew 8) 8.368 WE WON'T WORRY ABOUT ANY OF THAT
STUFF. 8.2007 F-C I'M WORRIED ABOUT THE GAS.
WE'RE PISSIN' OUT 15,000 POUNDS AN
HOUR. C-F DON'T WORRY. C-F DON'T
WORRY. F-C OK. C-F I'LL TELL YA WHEN TO
WORRY.
32
New Project Phase(June 2001 - )
DataFlight simulator data, but with a much
more constrained scenarioin which communication
skills are crucial
33
Flight Simulator Scenario

• Developed by Gerd Fahnenbruck, carried out by
  Lufthansa CityLine on Canadair Regional Jet
• Circling approach due to glide slope failure,
  runway has to be approached from unusual
  direction, a plane is blocking the runway, fly
  a go around
• Normal second approach, break down of glide
  slope requiring constant communication about
  altitude
• Take-off followed by double instrument failure
  requiring constant communication about
  conflicting values

34
Data, Transcripts, Analysis

• Video data (4 cameras) and audio data available
• 16 flights, average length 53 minutes
• Identification of three segments with
  different types of task load across crews.
• Careful transcriptions, including features
  of prosody, intonation, pauses
• Refined system of speech act classification,
  following Qualitative Development Analysis
  (T. Diegritz C. Fürst, Empirische
  Sprechhandlungsforschung. Erlangen 1999)

35
Example Steps of analysis (QVA)
36
Example Steps of analysis (QVA)
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.
37
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description

38
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description

After initiating the utterance with the particle
soCaptain utters a question with rising
intonation. Directly after this, as the captain
is looking at the instrument panel, he utters a
statement with strongly rising intonation.
39
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts

40
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts

• There are three subsequent illocutionary acts
• so
• wie weit sin mer
• flaps man wir twenty

41
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context

42
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context

Reading back Due to the instrument failure,
first officer informs captain that he can only
fly in stand-by mode. Captain agrees that it
would be best not to split up the tasksbut to
let first officer fly alone in stand-by
mode. Captain remarks that he can still help a
little bit.
43
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization

44
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization

We have made it clear how we will proceed. Let us
check the current state, to see how far we are
in respect to the planned approachwith the
instrument failure. I will go ahead and check the
flaps, which are 20.
45
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization
• Perlocutionary effects

46
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization
• Perlocutionary effects

First officer reacts with a confirming ja.
47
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization
• Perlocutionary effects
• Aspects of development

48
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization
• Perlocutionary effects
• Aspects of development

• Thematic progression Captain marks a
  thematic break by ending the previous sequence of
  clarification and going on to deal with the
  problem himself.
• Relational development Captain integrates
  first officer actively and cooperatively in
  the problem solving process, by asking
  himself and the first officer and by answering
  his own question.

49
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization
• Perlocutionary effects
• Aspects of development
• Assigning speech act type

50
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization
• Perlocutionary effects
• Aspects of development
• Assigning speech act type

So. STRUCTURING SIGNAL I indicate that I am
about to start a speech act and that the
previous discussion is over.
51
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization
• Perlocutionary effects
• Aspects of development
• Assigning speech act type

Wie weit sin mer? QUESTION(INFORMATION) I ask
you and myself about the current state of events.
52
Example Steps of analysis
so. wie weit sin mer? flaps ham wir twenty.So.
How far are we? As for the flaps we have 20.

• Description
• Division in speech acts
• Context
• Explicit interpretative verbalization
• Perlocutionary effects
• Aspects of development
• Assigning speech act type

Flaps ham wir twenty. ASSERT WITH EVIDENCE
(Feststellung) I state that the flaps are
twenty (and I have direct evidence for it)
53
Features of our Speech Act System

• Two further aspects of the speech act system we
  are developing
• - Speech Act Types
• - Dialogue Structure

54
Dialogue Segment
SA Line

• (4) 906 Fio ja
• (5) 906 Fio dann warten wa bis auf n
  intercept
• (6) 907 Cap weißte wir sollten alles
  raussetzen. gear down und flaps thirty.
• (7) 907 Cap dann ham wa das schon
• (8) 909 Cap is keen problem für uns
• (9) 909 Cap okay?
• (10) 911 Fio .hh hmm
• (11) 913 Fio ich bin mir nich sicher
  ehrlich gesacht.
• (12) 913 Fio aber gut. versuchn wa s mal

55
Examples Speech Act Types

• (4) Confirmation (BESTÄTIGUNG)
• (5) Suggestion (VORSCHLAG)
• (6) Suggestion (VORSCHLAG)
• (7) Giving reasons (BEGRÜNDUNG)
• (8) Statement (FESTSTELLUNG/BEHAUPTUNG)
• (9) Request for reply/confirmation
  (HÖRERRÜCKMELD. FORDERUNG)
• (10) Signal of doubt (ÜBERLEG/ ZWEIFEL ANZEIG)
• (11) Objecting/Doubting (ANZWEIFELN/EINWENDEN)
• (12) Conceding/Confirmation (ZUGESTEHEN/
  BESTÄTIGUNG)

56
STACK
8 Classes of Speech Act Types I.
Information II. Requests (direct/indirect) III.
Agreement/Negotiation IV. Dissent V. Question
types VI. Expressives VII. Interaction
markers VIII. Others
57
(No Transcript)
58
(No Transcript)
59
(No Transcript)
60
Hierarchical Structure of Speech Acts

•   
• Initial Reactive Reinitiative
• Accepting Problematic
  Rejective
• Statements Confirming Doubting Refuting
  Insisting
• (also Conceding Objecting
• Prognosis
• Report etc.)

Assertives
61
Structural Dialogue Analysis (Franke 1990)

• Analysis of sequential embedding in the
  communicative context
• Analysis of dialogue types (well defined
  sequences of speech acts)
• Basic unit minimal dialogue

62
Structural Dialogue Analysis (Franke 1990)

• With an initial speech act (ISA), speaker 1
  states a communicative goal
• Speaker 2 can accept or reject this goal
• For different ISAs there are well-defined sets
  of possible answers

63
Structural Dialogue Analysis (Franke 1990)

• Examples Sp2
• Sp1 answer 1 answer 2
• question answer refuse to answer
• assertion agree refute
• suggestion accept reject
• Further option counter-initiative speech act

64
Structural Dialogue Analysis (Franke 1990)

• Options for speaker 1 in the 3rd move (in
  response to a rejecting or counter-initiative
  second move)
• Retractive SA (RETSA) give up original goal
• Revised SA (REVSA) modify original goal
• Re-initiative SA (REISA) repeat the same goal

65
Dialogue Structure in our Data Segment

• SA structural type of SA
• 5 initial speech act (ISA)
• 6-9 counter-initiative speech act (GISA)
• 10-11 re-initiative speech act (REISA)
• 12 retractive speech act (RETSA)

66
Outlook on application of Dialogue Structures

• Assigning a structural type to speech acts
  enables us to represent a different dimension of
  dialogue
• The different ways that the crews speech acts
  relate to each other is an indicator of their
  communicative behavior
• This type of extended speech act theory allows
  us to make a quantitative comparison of aspects
  of dialogue structure with crew performance
  ratings