--------------------------------------------------------------------------- DCIEM SLEEP DEPRIVATION STUDY: THE DESIGN OF THE MAP TASK DIALOGUES --------------------------------------------------------------------------- I. Introduction II. The Sleep Deprivation Study II.A. Goal II.B. Design II.B.1. Drug Conditions II.B.2. Blocks and Tasks II.B.3. Subjects III. Materials III.A. Phonological Characteristics III.A.1. Phonological Modifications III.A.1.i. Master Features III.A.1.ii. Other Features III.A.2. Prosodic Structure III.B. Feature Types III.B.1. Introduction III.B.2. Contrast and Match III.B.3. Sharedness III.B.4. Odd-man-out III.C. Routes III.C.1. Description III.C.2. Routes and Master Features III.D. Quartets III.E. Assignment of Feature Names to Feature Types, Maps, and Quartets III.F. Lists and Diagnostics IV. Subject Conditions IV.A. Drug Condition IV.B. Subgroup V. The Full Design V.A. Design for Assigning Subjects to Master Maps, All Runs V.B. Schedule of Dialogues --------------------------------------------------------------------------- I. Introduction The dialogues on this CD-ROM are selected from the set collected during the DCIEM Sleep Deprivation Study (1994). In design, in detail and in purpose, the version of the map task used in this is based on the HCRC Map Task Corpus. Like its predecessor, the current version is intended to provide a common corpus for a vertical study of dialogue, generating material which can be discussed at levels from the acoustic to the sociolinguistic. All the parameters incorporated in the design are described in this document. The map task is a cooperative task involving two participants. The two speakers sit opposite one another and each has a map which the other cannot see. One speaker, the Instruction Giver, has a route marked on his/her map, while the other, the Instruction Follower, has no route. The speakers are told that their goal is to reproduce the Instruction Giver's route on the Instruction Follower's map. The maps are not identical and the speakers are told this explicitly at the beginning of their first session. It is, however, up to them to discover how the two maps differ. No restrictions are place on what either speaker can say. All dialogues in the present study were recorded via close-talking microphones, with one channel per speaker, on a Panasonic SV-3500 DAT recorder in quiet conditions. All participants took part in a number of sessions, and so gained experience with different maps. The maps themselves differ as a result of the systematic manipulation of the following design variables: 1. phonological characteristics of feature names 2. the extent to which features contrast or are shared between the maps The assignment of speakers to maps involves two further variables: 3. drug condition 4. subgroup size This document describes the design of the DCIEM Sleep Deprivation Study (Section II), the variables involved in the design of the Map Task materials (Section III), and the assignment of subjects to conditions (IV). II. The Sleep Deprivation Study II.A. Goal The Defense and Civil Institute of Environmental Medicine (Department of National Defense, Canada) has a history of studies into the effects of sleep deprivation. The present study was directed towards effects of drugs on performance decrements during sleep deprivation. It is known that amphetamines counter some decrements, but amphetamines have undesirable side-effects. For some time, the French Army had been making available to its soldiers on demand a drug called Modafinil, which had for some time been prescribed for narcolepsy, but which had not been tested for its effect on the performance of otherwise normal individuals deprived of natural sleep. Unlike amphetamines, Modafinil is known to have a wide ratio between useful and toxic doses. The present study was designed to determine whether Modafinil would work as well as amphetamines in reducing the performance decrements associated with sleep loss. The speech materials were collected by DCIEM as part of its participation in RSG-10 (NATO Panel 3 Research Study Group 10 (Automatic Speech Processing)), and designed by HCRC. For further details, consult Dr. M. M. Taylor (mmt@dciem.dnd.ca). II.B. Design II.B.1. Drug Conditions The study included 6 week-long replications of the same internal design, each called a RUN, each involving a new group of 6 subjects. In much of the documentation, and in file-internal headers, runs are numbered 110, 120, ... , 160. In some of the abbreviated forms, like short file names, runs are numbered 1, 2, ... , 6 Runs were assigned to drug treatments double blind: These were: 1. Modafinil (Runs 130 and 150) 2. Amphetamine (Runs 110 and 160) 3. Placebo (Runs 120 and 140) II.B.2. Blocks and Tasks An unusual feature of this study is that subjects' time was almost wholly occupied with the performance of various psychological tasks. These were scheduled in 6-hour BLOCKS of three 2-hour SESSIONS each. The task sequence was the same in each 6-hour block, but differed among the 2-hour blocks in some respects. The first day of the study was used for introduction to the tasks. Blocks began on the second day and continued throughout, except for sleep periods on second, fifth, and sixth nights. For all but two tasks, the Map Task and one other, subjects were isolated in separate rooms, where tasks were presented on a computer. The Map Task was performed in pairs under conditions described later in this document. The other `social' task involved group decision making. For this, all 6 subjects came together around a table to agree on the answers to trivia questions that they had previously attempted to answer by themselves. The group decision task was run at the end of the second of the three 2-hour sessions in each 6-hour block. A map task dialogue was produced by some pair of subjects as the final task of each 2-hour session. Each session finished with a rest period nominally of 15 minutes, but actually longer or shorter depending on how fast the subjects had performed the various tasks. As a result, subjects had an incentive to keep the dialogue short. If the session was one with a group decision task, the map dialogue was run after that task had been finished. Most other tasks took only a few minutes, or even less, and there were many of them. The following are examples: - Tracking tasks in which the subjects used a mouse to control some aspect of a screen display over a 50-second period. - Time estimation tasks, in which the subject had to press a key after a specified number of seconds had elapsed, or to indicate how many seconds passed between two events, or to indicate the current time-of-day (Subjects were denied access to radios, clocks, watches, or daylight). - Attention-splitting, in which they had to deal with four different kinds of things happening in the four quadrants of the screen. - Mood and physiological feelings questionnaires, - Rapid logical reasoning tasks (e.g. A is not to the left of B: A B. True or false), - Sequential subtraction (shown a number on the screen, subtract 9 from it, then 8 from the remainder, then 7... then 5, then 9 again). - Relative length judgment on left and right halves of a broken line, - Planning a route for two robots through a shopping market so that they could collect a given selection of goods in the minimum time. Drug doses were delivered after - Block 4, Session 2 - Block 9, Session 2 - Block 11, Session 1 This series approximated 18, 48, and 58 hours without sleep. II.B.3. Subjects The subjects were self-selected. All were military reservists, who volunteered when they heard about the project. Their ranks ranged from private to captain, and their ages from about 18 to 50 (see transcription file headers for biographical information). One subject dropped out just before Run 2. Only two subjects were female, though female participants had been invited subject to strict medical criteria: that they were not taking a contraceptive pill and that there was no possibility that they were pregnant. III. Materials III.A. Phonological Characteristics III.A.1. Phonological Modifications The maps draw upon FOUR phonological modification categories, or REDUCTION TYPES: 1. t-deletion 2. glottalisation 3. d-deletion 4. nasal assimilation Opportunities for reductions of these types characterize the names of landmarks. The use made of the landmarks subdivides them into MASTER FEATURES and OTHER KINDS OF FEATURES. III.A.1.i. Master Features Each map includes one or both of a potential pair of MASTER FEATURES (landmarks). For each reduction type, there is a different pair of master feature names, and each pair of master features appears on an equal number of maps in the corpus. The name pairs are as follows: Code Reduction-type MASTER FEATURE names 1 t-deletion east lake / west lake 2 glottalisation white mountain / slate mountain 3 d-deletion diamond mine / gold mine 4 nasal assimilation crane bay / green bay The description of CONTRAST/MATCH given below explains how maps come to have one or both of their master features. III.A.1.ii. Other Features In addition to master features, landmark names offering sites for the four categories of reduction type occur on the maps OTHER FEATURE TYPES, which will be described in more detail below. Each map contains at least one example of each reduction type. III.A.2. Prosodic Structure Each map contains at least one example of each of two POLYSYLLABIC CATEGORIES: 1. initial STRONG-WEAK (eg "buffalo") 2. initial WEAK-STRONG (eg "baboons") III.B. Feature Types III.B.1. Introduction The maps include labelled drawings depicting a number of landmarks or features, arranged on the page in a systematic way. The Giver's and Follower's maps were carefully constructed to include features which differed along a number of dimensions: 1. CONTRAST 2. SHAREDNESS 3. ODD-MAN-OUT Additional INCIDENTAL FEATURES were included for lexical variety. III.B.2. Contrast and Match Over the maps in the design, the pairs of master features listed above appear in a balanced set of CONTRAST conditions. If there is a contrast (+), then both members of the master feature pair are on the Instruction Giver's map, and the Instruction Giver has an opportunity to subject members of the pair to contrastive stress. If there is a match (+), the Instruction Follower's map matches the Instruction Giver's in respect of contrast. Thus either Giver or Follower or both or neither may have the pair of master features. There are equal numbers of Giver-Follower map pairs of each of the following types: ++ Both maps have both members of their master feature pair. +- The Giver's has both, the Follower's only one. -+ Both maps have only one member of the pair. -- The Giver's has one, the Follower's both. III.B.3. Sharedness Other features on each map belong to a number of categories dependent on whether the name, the illustration, or the number of instances of the feature are the same on both maps in a pair. Examples of the following categories of sharedness were included on each map: COMMON FEATURE a feature which is common to both the Instruction Giver and Follower's map; ie the same drawing occurring with the same name in the same location on the map. NAME CHANGE a feature that is common to both maps (same drawing, same location) but which is named differently on the two maps; eg., where the Instruction Giver has "white water" the Instruction Follower might have "rapids". ABSENT/PRESENT a feature that is present on one speaker's map but not the other's TWO-TO-ONE (2:1) a feature which the Giver has two of, one of which is relevant to the route and one irrelevant. The Follower only has the irrelevant feature. For each map, each one of these sharedness types occurs with a different one of the four reduction types. III.B.4. Odd-man-out Feature names were chosen for meaning as well as for the sound of their names. A SCENARIO was devised for each map and features chosen to fit in with this stereotypical location. For example, there might be a "Wild West" scenario, with "Apache camp", "canoes", "buffalo", "gold mine" and "cavalry" as landmark names. One feature, the ODD-MAN-OUT, would be alien to this scenario, like a "nuclear test site" in the "Wild West". III.C. Routes III.C.1. Description Four different routes were constructed for the maps. To help ensure that the routes differed sufficiently, random number tables were used to generate grid co-ordinates which determined the positions of the major features (those involving a potential phonological modification) on the page. A route was then drawn around the features observing the following criteria: 1. The route starts at a shared, or common, feature 2. The route finishes at a common feature 3. Intermediate landmarks along the route alternate between common features and those that differ in some way (see above) 4. There are at least two features which only appear on the Giver's map, and two features which only the Follower has. III.C.2. Routes and Master Features The four routes are associated with particular master features. For example, the master feature "east lake" always occurs in the same location on any map in which it appears, and these maps all share the same route. For this reason, routes can be assigned the number given to the phonological reduction type of the master feature. Thus the route associated with the master feature "east lake" is route number 1 as "east lake" is the t-deletion master feature. III.D. Quartets Sixteen maps were constructed in the manner described above. Each has a unique ROUTE(master feature) X CONTRAST/MATCH combination. Four different QUARTETS of maps were created using a latin square to ensure that each contained one example of each contrast/match condition and of each route, but that no ROUTE (master feature) X CONTRAST/MATCH combination occurred more than once. In the HCRC Map Task Corpus, the 16 maps (4 for each quartet) were allocated as shown below. In the DCIEM study, the first quartet was not used. Quartet Contrast/Match/Route Qrt1 ++1 +-2 -+3 --4 Qrt2 ++4 +-1 -+2 --3 Qrt3 ++3 +-4 -+1 --2 Qrt4 ++2 +-3 -+4 --1 where +CONTRAST = Instruction Giver's map contains contrasting master features (eg "east lake" and "west lake") -CONTRAST = Instruction Giver's map contains only one member of the master feature pair (eg "east lake") +MATCH = Instruction Follower's map matches Giver's in contrast, so if Giver has both lakes so does Follower, but if Giver has only one then Follower only has one -MATCH = Instruction Follower's map mismatches Giver's in contrast, so if Giver has two lakes Follower has one, if Giver has one, Follower has two ROUTE 1 = associated with T-DELETION master feature ROUTE 2 = associated with GLOTTALISATION master feature ROUTE 3 = associated with D-DELETION master feature ROUTE 4 = associated with NASAL ASSIMILATION master feature III.E. Assignment of Feature Names to Feature Types, Maps, and Quartets The tables below indicate which landmarks were used for each feature type in each map. Names marked with an asterisk differ from the corresponding items in the HCRC Map Task Corpus, because the originals were likely to be unfamiliar to Canadians. =========================================================================== Map TYPE OF SHAREDNESS (master features) 2:1 ABSENT/ NAME COMMON ODD MAN OUT PRESENT CHANGE =========================================================================== Quartet 2 ++4 3 4 1 2 3 crane bay farmed iron forked pirate computer green bay land bridge stream/ ship controlled *babbling brook sub +-1 2 1 4 3 2 east lake great popular farmer's ruined rocket west lake viewpoint tourist gate/ monastery launch pad spot broken gate -+2 1 2 3 4 1 white mountain lost straight ancient fallen *biggest slate mountain steps river ruins/ pillars furniture ruined city store --3 4 3 2 1 4 diamond mine stone manned white water/ rift swan pond gold mine creek fort rapids valley =========================================================================== Quartet 3 ++3 2 1 4 3 2 diamond mine great parched indian carved trout gold mine rock river bed country stones farm +-4 1 2 3 4 1 crane bay vast white- reclaimed seven crashed green bay meadow washed fields beeches spaceship cottage -+1 4 3 2 1 4 east lake train privately granite site of lion west lake crossing owned quarry forest country fields fire --2 3 4 1 2 3 white mountain poisoned lemon crest falls remote cobbled slate mountain stream grove village street =========================================================================== Quartet 4 ++2 4 3 2 1 4 white mountain golden submerged secret extinct *Elizabethan slate mountain beach rocks valley volcano barn +-3 3 4 1 2 3 diamond mine field overgrown highest great *closed gold mine station gully viewpoint lake warehouse -+4 2 1 4 3 2 crane bay boat washed pine pebbled coconut green bay house stones grove shore palm --1 1 2 3 4 1 east lake parked flight *abandoned telephone thatched west lake van museum monastery box mud hut =========================================================================== NB Due to an error there is no change of name on the maps in Quartets 3 and 4 III.F. Lists and Diagnostics At the start of each Map Task session subjects were asked to read four accent diagnostic sentences taken from Barry, Hoequist, and Nolan (1989): 1. After tea father fed the cat. 2. Father hid that awful cart at the top of the park. 3. Father cooked two of the puddings in butter. 4. Father bought a lot of cloth. After completing the Map Task conversations subjects read a wordlist containing all the landmark names on the maps used by those subjects. IV. Subject Conditions Unlike the HCRC Map Task Corpus, which systematically varied speaker familiarity and eye contact, the current study made no attempt to control the former and held the latter constant. All subjects performed facing one another across of a pair of tables, but an office space divider prevented them seeing one another. Instead the variables classifying subjects were: 1. DRUG CONDITION 2. SUBGROUP SIZE IV.A. Drug Condition All subjects in a run received the same drug treatment. For the assignment of runs to treatments, see Section II.B.2 above. IV.B. Subgroup Within each run subjects were divided into two subgroups: 1. a FOURSOME 2. a PAIR. In each case, the members of the foursome are called subjects 1-4 (or, more specifically, 111-114, 121-124, etc.). They followed the pattern of partnering which was used in the HCRC Map Task Corpus. This was based on a foursome contributing 8 dialogues. Following the original terminology, we continue to call this set of 8 dialogues a QUAD, whether or not 4 participants were involved. Over each 8 dialogues, each subject served as Instruction Giver twice, using the same map, and as Instruction Follower twice using different maps. Each subject worked with two of the other three subjects in his/her foursome. The pair includes subjects 5 and 6 (115-116, etc.). They worked only with one another, alternating in Giver and Follower roles. Over each set of 8 dialogues (each Quad) they encountered each map twice, once as Giver and once as Follower. The order of maps was the same for both subgroups. But because only one dialogue was run per session and pairs were given half as many sessions as foursomes, the foursomes collectively completed 24 dialogues per run, or 12 per speaker, and pairs only 12 dialogues per run, but once more 12 per speaker. V. The Full Design V.A. Design for Assigning Subjects to Master Maps, All Runs ------------------------------------------------------------------------- Quad Subjects -------- -------------------------------- Master feature 2 3 4 Foursome Pair treatment --------- --------------- --------------- ------------- Map/Master Conversation Giver Follower Giver Follower Con- Match Reduction Number trast ------------------------------------------------------------------------- 4 3 2 1 1 3 5 6 + + 1 4 3 2 4 2 6 5 + - 2 1 4 3 2 1 5 6 - + 3 2 1 4 3 4 6 5 - - 2 1 4 5 2 4 6 5 - + 3 2 1 6 3 1 5 6 - - 4 3 2 7 1 2 6 5 + + 1 4 3 8 4 3 5 6 + - ------------------------------------------------------------------------- Key: Subject Codes: the numbers 1 - 6 were assigned randomly to participants in a run. Reduction: 1 = possible t-deletion (east/west lake) 2 = possible glottalization (white/slate mountain) 3 = possible d-deletion (diamond/gold mine) 4 = possible nasal assimilation (green/crane bay) Contrast: + both members of pair present on Giver's map - only 1 member of pair present on Giver's map Match: + Follower's map matches contrast on Giver's - Follower's map mismatches contrast on Giver's V.B. Schedule of Dialogues This table indicates which dialogues were run in which blocks and which subjects participated. It also indicates the filename used for materials relating to these dialogues. ------------------------------------------------------------------------- Time Day Sun Mon Tues Wed Thu Fri ---------------------------------------------------------------- 0:00 SLEEP SLEEP b4s3 b8s3 SLEEP -1:30 q3c4 q3c8 g6f5 g5f6 d15 d27 x15q3c4p.y x27q3c8p.y 2:00 SLEEP SLEEP b5s1 b9s1 SLEEP -3:30 q3c1 q4c1 g1f3 g1f3 d16 d28 x16q3c1f.y x28q4c1f.y 4:00 SLEEP SLEEP b5s2 b9s2 SLEEP -5:30 q3c2 q4c2 g4c2 g4c2 d17 d29 x17q3c2f.y x29q4c2f.y >DRUG< 6:00 b5s3 b9s3 SLEEP -7:30 q3c5 q4c1 g6f5 g5f6 d18 d30 x18q3c5p.y x30q4c1p.y 8:00 b2s1 b6s1 b10s1 SLEEP -9:30 q2c3 q3c3 q4c3 g2f1 g2f1 g2f1 d7 d19 d31 x7q2c3f.y x19q3c3f.y x31q4c3f.y 10:00 PRACTICE b2s2 b6s2 b10s2 SLEEP -11:30 D'LOGS q2c4 q3c4 q4c4 d1-d3 g3f4 g3f4 g3f4 d8 d20 d32 x8q2c4f.y x20q3c4f.y x32q4c4f.y 12:00 DEMO b1s1 b2s3 b6s3 b10s3 b12s1 -13:30 q2c1 q3c2 q3c6 q4c2 q4c7 g1f3 g6f5 g5f6 g6f5 g1f2 d4 d9 d21 d33 d37 x4q2c1f.y x9q3c2p.y x21q3c6p.y x33q4c2p.y x37q4c7f.y * 14:00 b1s2 b3s1 b7s1 b11s1 b12s2 -15:30 q2c2 q2c5 q3c5 q4c5 q4c8 g4f2 g2f4 g2f4 g2f4 g4f3 d5 d10 d22 d34 d38 x5q2c2f.y x10q2c5f.y x22q3c5f.y x34q4c5f.y x38q4c8f.y * * * >DRUG< 16:00 b1s3 b3s2 b7s2 b11s2 b12s3 -17:30 q3c1 q2c6 q3c6 q4c6 q4c4 g5f6 g3f1 g3f1 g3f1 g6f5 d6 d11 d23 d35 d39 x6q3c1p.y x11q2c6f.y x23q3c6f.y x35q4c6f.y x39q4c4p.y * * * 18:00 b3s3 b7s3 b11s3 WORD -19:30 q3c3 q3c7 q4c3 LISTS g5f6 g6f5 g5f6 d12 d24 d36 x12q3c3p.y x24q3c7p.y x36q4c3p.y * * 20:00 b4s1 b8s1 -21:30 q2c7 q3c7 g1f2 g1f2 d13 d25 x13q2c7f.y x25q3cp.y 22:00 SLEEP SLEEP b4s2 b8s2 SLEEP SLEEP -23:30 q2c8 q3c8 g4f3 g4f3 d14 d26 x14q2c8f.y x26q3c8f.y >DRUG< ------------------------------------------------------------------------- ABBREVIATIONS: Block key: b block (1-12) s session (1-3) Quad key: q quad (2-4) c conversation within quad (1-8) Participant key: g giver subject number (1-6) f follower subject number (1-6) Dialogue number: d dialogue within whole run (1-39) Filename key: x run of the experiment (1-6) f foursome (some pair of subjects within 1-4) p pair (subjects 5 and 6) y a 3-character suffix for filetype (trn, ses, etc.) Table of Contents