Discourse Analysis of Crew Cooperation
Hideyuki ANDO (The University of Tokyo, Japan)
Hiroyuki YAMATO (The University of Tokyo, Japan)
Takeo KOYAMA (Japan Marine Science, Inc.)
Ryo KAKUTA (The University of Tokyo, Japan)
Hisanori NISHIDA (Aristotles, Inc.)
Masakazu ENOMOTO (The University of Tokyo, Japan)
Abstract: For reducing human errors in navigating ships, crews on a bridge must cooperate. BRM (Bridge Resource Management) are techniques to achieve effective cooperation on a bridge. However, how crews cooperate has not been well understood yet. To understand crews' cooperation, we developed a system CORAS (COllaboration Record and Analyze System). CORAS supports discourse analysis of crew interactions by recording video, audio and simulator logs, and allow users to effectively retrieve those records. We conducted discourse analysis on two BRM trainings by using CORAS. By quantitative discourse analysis, communication flows, distribution of exchanged information and communication types, roles of crews are revealed. To understand cooperation in detail, qualitative discourse analysis is conducted. Sequential analysis shows several patterns of team cooperative work on a bridge.
For reducing human errors in navigating ships, team members on bridge must cooperate by exchanging information, checking other crews' errors and delegating tasks. Those cooperation techniques are called as BRM (Bridge Resource Management) .
BRM has acquired attentions as a way to prevent incidents. However, deeper understandings of team cooperation are required to improve evaluation methods in training, to revise training program and to design manuals.
This paper presents discourse analysis methods to understand the way of cooperation in a bridge and also introduces a system to support the analysis.
2. TEAM SITUATION AWARENESS,DECISION MAKING AND ACTION
Tasks on a bridge are similar to those in a aircraft cockpit and in a control room of railways, where small teams cooperate, aware surrounding situations and make decisions in uncertain conditions and under time pressures. Decision making processes in those real settings are treated ~s Naturalistic Decision Making (NDM) . We apply Endsley's NDM model , which consists of situation awareness, decision making and action, to describe the team decision making process for organizing tasks and cooperation patterns on a bridge. Fig. I shows the team decision making model.
Fig. 1 A team decision making model to understand bridge cooperative work
3. ANALYSIS METHOD
To analyze how team members cooperate in duties, we build a system to record team practices in BRM trainings. The system is called as CORAS (COllaboration Record and Analyze System). CORAS has functions to record multiple types of time-series data, such as video, audio and simulator logs.
Fig.2 Work recording by video and audio
Fig. 2 shows the recording environment for video and audio in a simulator room. Fixed camera records actions of crews in front. Hand-held camera records actions, displays of instruments, the chart table and projected situations. The size of simulator room is about 5-meter-square and one capacitor microphone can record all conversations between crews in the room. Video and audio data are imported via IEEE1395 to PC and are converted to MPEGI file in real time by using a MPEG1/2 hardware encoder. Simulator logs are imported from the simulator server and stored in tables of a relational database.
The user interface of CORAS is shown in Fig 3. A user can replay recorded data by using the time controller at top-left comer, which has a slider-bar to select replay position and an input box to specify exact seconds to rewind or fast-forward.
Addition to the time controller, a user can add annotation titles and texts on any time windows of work records. By using those annotations, user can retrieve the exact time frame. This function is used for recursive replays and share information between several users.
CORAS replays video, audio and simulator logs. On the right form, a user can select a range of displaying data and can read ship information, such as position, bearing, distance, speed, TCPA (Time before Closest Point of Approach) and DCPA (Distance to Closest Point of Approach) by selecting a ship by a pointer.
Fig.3 The user interface of CORAS
Addition to video, audio and simulator logs, we transcribe discourses in the bridge by observing work records stored in CORAS. Interactions between crews play major roles in team collaboration and have rich information to understand the cooperative operations' status on the bridge. Transcripts are stored in a spreadsheet for facilitating later analysis.
To understand team cooperation, we apply discourse analysis. Discourse analysis aims at understanding meaning of a discourse. In this paper, we analyze discourse by the following methods.
・Quantitative analysis Structuralize discourses and categorize them into several codes. Quantitative data is acquired by counting frequency of each code appearances.
・Qualitative analysis - Observe discourses and find typical methods or patterns. Hutchins showed a good practice of qualitative analysis of team cooperation on a bridge .
・Sequential discourse analysis - Is a type of quantitative discourse analysis. It counts time dimension of discourse and find patterns or structures in a discourse.
4 DISCOURSE STRUCTURE ON BRIDGE
For quantitative analysis, we need to define a discourse structure for bridge conversations. Fig.4 shows a typical example of a discourse on the bridge.
In this example, Captain, 2nd officer (2/O) and Quarter Master (Q/M) are cooperating. Firstly. Q/M reports the course of the ship to Captain. Captain asks the course of the other ship, Yesterday, to 2/O. 2/O is responsible for lookout by using ARPA (Automatic Radar Plotting Aids). 2/O finds the course of Yesterday on ARPA and reports it to Captain. Captain also asks distance to Yesterday and 2/O reports it to Captain with speed information. Then Captain decides to change course and order it to Q/M.
Fig.4 An example of discourse among crews
From the series of communication between crews, it is observable that there exists a strict organization on the bridge. Captain makes a decision and other crews help Captain to recognize necessary information and perform actions based on Captain's decisions. Role assignments of crews are clear. In this case, 2/O is responsible for lookout, Q/M is responsible for helm.
Comparing to a discourse in ordinal situations, such as daily conversation, a discourse on the bridges seem to have following characteristics.
・Basically, topics in discourses are related to the limited number of operation duties.
・Patterns of utterances of each speaker largely depend on role assignments.
・Structures of discourse are clear. Lots of utterances are consisting of pairs, such as order-repeat and report-acknowledge.
By using the third characteristics, we can define a discourse structure for the bridge collaboration. Fig.5 shows a hierarchical discourse structure. This hierarchy consists of 5 levels, dialog, session, sequence, exchange and utterance. Among those levels, exchange is a minimal set of utterances and corresponds to a unit of work, such as course recognition and change speed.
Fig.5 Hierarchical discourse structure
Each utterance consists of timestamp, speaker, listener and content.
We define a set of codes, which is assigned to exchange level structure. The codes, shown in Fig. 6, consist of work information categories and communication type categories. Those codes are defined, after reviewing discourse transcripts in two BRM trainings.
Fig.6 Category of codes for exchange level structure
Category of information
1. another ship - bearing, course, side, name, exsistence, speed, cpa, shiptype, radar, binocular, losttarget
Category of communication type
By utilizing those codes and discourse structure, the following information can be retrieved from transcript.
・Patterns of exchanged information
・Role of each speaker