Upload
rui-sakaida
View
112
Download
0
Embed Size (px)
Citation preview
Rui Sakaida Fumitoshi Kato Masaki Suwa (Keio University)
HOW DO WE TALK IN TABLE COOKING?
October 28th, 2013 MiMI 2013
A Little Strange Situation: Why aren’t They Embarrassed?
• However, ü they are not embarrassed… ü they are enjoying themselves…
• Why? • They are cooking “monja-‐yaki” together!
3 close friends sitting face-‐to-‐face around a small table
Several times of silence over 10 sec
Nobody repairs overlaps of utterances
Hypothesis: Table Cooking Generates Overlaps and Silence
• In general ü Overlaps and gaps are designed to be as few as possible Ø Transitions with no gap and no overlap are common
(Sacks et al., 1974)
• In table cooking ü We observed a lot of overlaps and silence ü Is special order of interaction allowed?
Japanese Table Cooking Style • The Japanese often cook and eat dishes on a table
nabe-‐ryori yakiniku temaki-‐zushi (one-‐pot meal) (grilled meat) (do-‐it-‐yourself sushi rolls)
• Table cooking makes social relationships among the participants ü What aspects of table cooking contribute to that effect?
Monja-‐yaki (Monja) • Thin and lax pancake with various ingredients
• Why monja-‐yaki?: intersting to observe in two ways ü Everyone can take part in cooking Ø Cf. Nabe-‐ryori (one-‐pot meal): One participant can monopolize cooking
because only one ladle is often served (we call him/her “nabe-‐bugyo (chair-‐person)”)
ü Difficult to cook Ø Process of cooking is complicated Ø More skilled participants may tell how to cook to less skilled participants
monja-‐yaki
Multiparty Interaction
• The next speaker is not clear • Gaze exchanges can realize
smooth turn-‐taking ü A hearer gazed at by the current
speaker is likely to be the next speaker (Enomoto, 2011)
What is Multiparty Interaction in Table Cooking?
hearer speaker hearer speaker
?
Two-‐party Interaction • The current hearer will
always be the next speaker
What is Multiparty Interaction in Table Cooking? Multiparty Interaction in Table Cooking
• We must gaze at dishes or cooking tools ü Smooth turn-‐takings can be hampered ü Troublesome turn-‐takings generate overlaps and silence?
Why and How Does Table Cooking Generate Overlaps and Silence?
• Overlaps and silence may be allowed because of the unique characteristics of monja-‐yaki
Data: Three-‐party Table Cooking of Monja • Three-‐party conversation in a monja restaurant ü 1st author (S) invited 2 friends (U and H) • Recorded with 2 video cameras • What topics to talk are not given
The 27th Annual Conference of the Japanese Society for Artificial Intelligence, 2013
- 2 -
>ŀȽDžƜ�ąȥƬɗɕȽȠ�eʐ�ƤŇȽǰƵɏİȕȥ
ǷāǹŞŗȮȤȿȽȠȥʐƁ�ƩƜɀŦÜȺɁʐǍŔȵȤ
ɒȻȠȷȹɥʃʆɵɤʏɨʇʍɀĸ×Pȥ�¡ȬɕȴȻȠȡ Ü
ɁȽȠʑUfǍŔɀ�ȾȣȠȹɁʐ\�Ɯ9oȺUJȮȹù
Ŕɘ�áȬȲɔȻȠȡʐþŰȤȸB ȽťŠȥ:ćȬɕȹȠɔʑ
ǍŔɀǡƬȾɥʃʆɵɤʏɨʇʍȥòǭȬɕȹɍʐǍŔǢƬȻȠ
ȡ�ƙgAɀɍȻȾʐŸÀɀ±�ȥǂȬɕɔɀȤɍȮɕȽȠʑ ȳɍȳɍɥʃʆɵɤʏɨʇʍȾȣȩɔŞDžɀǰƵɏİȕɁʐ�
ƥȾɁĬɉȱȬɏȖȗɘŗɋȜŞŗȰɆȦȺȽȠœǒȝȻȮȹ
ìȢɒɕȥȶȺȟɔȥʐȳȡȻɁǼɒȽȠʑŧĝɀǰƵɁ�Džɀ
� ÜɘǯáȰɔȺȟɖȡȮ1ʐDžƜ�ąȾȣȠȹŞDžĉƶƊ
ȥǰƵȰɔœǒɁʐěȮȹ>ŀȽDžƜ�ąȻDŽ�Ȭɕɔ[ĜĊ 03]ʑɉȴƺ¥ÃɀȓȠ\�Ɯf�ȾȣȠȹɁʐǤÃȾĬǭɓȰɔȫȻȽȨʐ�«ɀǰƵɏİȕɘǂȮeȡȤɍȮɕȽȠʑȳɀ
�eʐƺ¥ÃȥǰƵɏİȕȻȮȹƯ@ȮȹȠɔȻɍƛȢɒɕɔʑ UfǍŔ�ȇȾȣȠȹɍʐŞDžɀǰƵɏİȕȥÔȱȮɍi
�ŠȽËȉɘ�ȢȹȠɔȻɁÜȯɒɕȽȠʑŞDžɀǰƵȥǍŔ
ȈȻɍŧɉȷȹ�Džɀţɓ�ȥɓɘÜȯȬȲʐİȕȥƓȠȹɍǍ
ŔɀĞ�ɘ9oȺƻ¦ȮȹȠɔȫȻȾɑɓʐ�Džɀ� Üȥ
ðȽɗɕȽȠʑ ĊŻȺɁʐUfǍŔɘ�ȡȏTȾȣȠȹʐǍŔȻȆǍŔ
ȺŞDžɀǰƵȻİȕɀŞŗāǹȣɑɄȊÃȥȠȤȾŝȽɔȤ
ĘǃȰɔʑɉȴĞțȽ:f"ęɀȽȤȺɍǍŔƬŅȾōćɀ
2hɘƷ@ȰȴɌȾʐǍŔƬŅ��ɀ:f"ęȻȮȹʐȏTȾ
ȹ�ȸɀʄɵʆʏƯɘ_ɓxəȺʄɵʆʏɘǨɄȽȥɒŗǖȰɔ
�DžȻɍĪǚȰɔʑ
4.
4.1 ���� ŗĸȾĕȭȮȴŖ��Ⱥɀȏ �Džɘ œȰɔȴɌʐƁ�
ƩƜȥ ŗĸȺƺ�ɀȟɔ\�Ɯ 2 g(HʐU)Ȼɀȏ ɀ�ɘǁ�ȮʐƁ�ƩƜȥ\�ƻ¦ȰɔȻȻɍȾʐȏTɀĀ3ȻȈ�
ɘƿǵȮ ÎȾAēȮȴʑ3 ��DžȺɁ�DžȥƵöɀɣʋʏʀȾAưȰɔaơ×ȥ�Ȩʐȏ �DžɀAēȾǧȮȹȠɔ[Ħ³ 11]ʑƤŇȽ�Džɘ œȰɔȴɌʐ�DžɀɱʏʂȽȼɀôŲɁȮȹȠȽȠʑ ŗĸɀȏ �ȇɘȺȦɔǼɓÖ Ⱦ?œȰɔȴɌʐŵ�³
Ũƫı¸=Ⱦ zȰɔȣ�ɊňȦÁȾȹ�Džɘ^ǵȮȴʑU
fȺǍŔȰɔŐſȻȮȹʐȣ�ɊňȦʒɍəȯɎňȦɘířȮȴʑ
HǞȮȴȻȣɓʐǶùŔɏňƞȽȼɀ�eʐō�ɀ\�ƜȥȠ
ɗɐɔǶ�ƬȻPȮȹǍŔƬŅɘŏWȮȹȮɉȡȣȳɕȥȟɔʑ
ȳɕȾ¨Ȯȹȣ�ɊňȦɏɍəȯɎňȦɁʐ9\�ƜȥǍŔȾ
\�Ȱɔ!{ɘ�ȢɔʑōȾɍəȯɎňȦȺɁʐ\�Ɯɀ��
ȥ;ČɘdzĒȾƨȻȰȻfāȾʐ�\�Ɯȥʁʉɘ$řȮȹƨ
ȻȬɕȴ;ČɘBɓFɋȽȼʐƵö\�ƜȥUfȮȹǍŔȰɔ
�ȇȥ�Ȩƻ¦ȬɕɔʑɉȴǍŔȃÿÃȥȓȠȴɌʐǍŔƎȒ
Ⱦ¶ɀȟɔ\�Ɯf�ȺʐôȢeȠɏǍŔĠɀǑļȥȮɂȮ
ɂŞŗȰɔʑȫɕɁʐĊůžȾȣȩɔUfǍŔɀAē¨ǒȻȮ
ȹŔÚŠȽÊÆȺȟɔʑ �Dž^ǵȺɁƁ�ƩƜ S Ȼ HȥȁɓeȠʐHɀĥȇȾ Uȥ
ŪºȮȴʑ ȒŖ�ɀȏTȾɁǷûÊɀdzĒʐǍŔřʁʉ 2ĊɀɇȤʐd\�ƜɀHȾ_ɓŢʐ,�řʁʉ(�ª 2 Ċȱȸ)ȥřÛȬɕȹȠɔʑȬɒȾȏTɀƢɀ HʐU /ȾɁɬʏɪ 3źʐ
1 �$�!&"��(� '�(%��)����*�#��"�����!&�(�������'��*���(%
��
ȔƇʐȅɀɓʐɦʉɮIJʐǮIJʐS /ȾɁʂʈɶʏɫʐȤɒȮʂʈɶʏɫɀ£tȥƗȤɕȹȠɔʑȏTȺî&ȬɕȴŐſɁʐɍə
ȯɎňȦ 2 nʐȣ�ɊňȦ 1 nʐɬɾɳɴʊʍɢ 3 gAȺȟɔʑŐſɀdžƋȽ=£Ɂʐǟ¼ɀȏ ȻfĞȾʐ\�Ɯ9oȺʄɵʆ
ʏƯɘƷȽȥɒǨçȮȴʑ 2 bɀɲɩɭʋɼɲɟɠʄʉȻ 3 bɀ IC ʌɥʏɮʏȣɑɄɽ
ʍʂɜɢɘ$řȮʐȈ�]ɄĀ3Ⱦȹ^ǵȮȴ(y 1)ʑɼɲɟɠʄʉɁ 40.5mmș59mmș30mm ȻȠȡʐĚɌȹª�ɀ GoProųƴ HERO3 ɘċ�ȾǁƗȮʐ\�ƜȾ^ǵȾɑɔǤÃɀƔÈɘ�ȢȽȠɑȡǭÞȮȴʑIC ʌɥʏɮʏɁɦʍʈʏųƴ ICR-PS004MʐɽʍʂɜɢɁɬɵʏųƴ ECM-CS3 ɘ$řȮʐd\�ƜɀƠ5ȾƱŪȮȹþŬȽȈ�ɀǵȈɘɌȭȮȴʑ
y 1: 2bɀɼɲɟɠʄʉȾɑɓ^ǵȮȴĀ3ɀɡʅʀɯʅś3
(µɁäHȤɒ\�Ɯ HʐS(Ɓ�ƩƜ)ʐcɁ\�Ɯ U)
4.2 ����9=7 Ư 1: Aē¨ǒɲʏɭěƶ
(1)ʄɵʆʏǨɄ S(Ɓ�ƩƜ) H U ƕŞDžāǹ(ŷ) 46.75 63.40 47.82 ƕŞDžuö(u) 47 66 45
1ŞDžȟȴɓɀ¾}āǹ(ŷ) 0.99 0.96 1.06 9oɀƕŞDžāǹȾWɌɔ
ŞDžāǹ(%) 29.59 40.14 30.27
(2)ùŔDŪÍȶ S(Ɓ�ƩƜ) H U ƕŞDžāǹ(ŷ) 43.80 23.29 42.26 ƕŞDžuö(u) 47 19 36
1ŞDžȟȴɓɀ¾}āǹ(ŷ) 0.93 1.23 1.17 9oɀƕŞDžāǹȾWɌɔ ŞDžāǹɀIe(%) 40.06 21.30 38.65
(3)ɍəȯɎňȦǍŔ S(Ɓ�ƩƜ) H U ƕŞDžāǹ(ŷ) 44.17 53.20 76.59 ƕŞDžuö(u) 47 52 65
1ŞDžȟȴɓɀ¾}āǹ(ŷ) 0.94 1.02 1.18 9oɀƕŞDžāǹȾWɌɔ ŞDžāǹɀIe(%) 25.39 30.58 44.03
^ǵȬɕȴ�Džɲʏɭ(ƽ 115A 14ŷ)ɀȡȶʐ(1)ʄɵʆʏǨ
Ʉʐ(2)ùŔȥǣɂɕȹȨɔɉȺɀÍȶāǹʐ(3)ɍəȯɎňȦɀǍŔȻȠȡŝȽɔ 3 ɾɞʏɫɘAē¨ǒƈxȻȰɔʑdɾɞʏɫɀAē¨ǒāǹɁʐ(1)234.85 ŷʐ(2)158.82 ŷʐ(3)412.93ŷȺȟɔʑdɾɞʏɫȾȣȩɔd\�ƜɀƕŞDžāǹʐƕŞDž
uöʐ1 ŞDžȟȴɓɀ¾}āǹʐ9oɀƕŞDžāǹȾWɌɔd\�ƜɀŞDžāǹɀIeɘƯ 1ȾŲȰʑ
4.3 5:8=6;< Ā3ɲʏɭAēɬɾɳɝɞɚ ELAN ɘ$řȮȹĀ3ȻȈ�ɘ
fĈȮʐŞDžȻǍŔƬŅɀćņɘɚɷɱʏɨʇʍȮȴʑǍŔƬŅ
ɀɚɷɱʏɨʇʍȾȣȠȹɁʐ[Kendon 04]ʐ[Ƌȑ 08]ɀɩɞɪɯʅʏAēV�ȾȣȩɔȜĿ1ȝȤɒȜÒ»ȝɉȺɘǍŔȻDŽ
�ȮȹȠɔʑ
Data Excerption • The data was divided into 3 phases (1) Seeing the menu and deciding what to eat (2) Waiting for the dishes to arrive (3) Cooking monja-‐yaki
Phase Bodily acts Participants gaze at
(1) Deciding what to eat Seeing the menu Non-‐verbal communications
Menu The other participants
(2) Waiting for the dishes Non-‐verbal communications The other participants
(3) Cooking monja-‐yaki Cooking Non-‐verbal communications
Dishes Cooking tools The other participants
Quantitative / Qualitative Analysis • The data was analyzed ü Quantitatively Ø Calculating hours / times of each participants’ utterances,
overlaps and silence
ü Qualitatively Ø Applying conversation analysis (CA), including gaze
directions
• We compare 3 phases quantitatively / qualitatively ü From a viewpoint of overlaps and silences
ANALYSIS OF OVERLAPS QUANTITATIVE ANALYSIS QUALITATIVE ANALYSIS
ANALYSIS OF SILENCE QUANTITATIVE ANALYSIS
Overlaps by Each Participant
Fig. The length and the number of times of overlaps among total length of utterances by each participant (%)
• In (1) and (3), overlaps of every participant are more frequent than in (2) (Except the length of overlaps of H) • Why do overlaps increase while cooking or looking at a menu? ü The reason will be indicated by means of CA
0
20
40
60
80
S H U
(1) Deciding what to eat
Length of overlaps (%) Times of overlaps (%)
0
20
40
60
80
S H U
(2) Waiting for the dishes
Length of overlaps (%) Times of overlaps (%)
0
20
40
60
80
S H U
(3) Cooking monja-yaki
Length of overlaps (%) Times of overlaps (%)
Qualitative Analysis by Means of CA • Order of overlaps in table cooking is indicated in two aspects: (1) Accidental overlaps are not always repaired in cooking (2) Co-‐telling of how to cook causes utterances to overlap
2 types of overlap
1. Accidental overlap • Simultaneous Start (and End)
ü Speakers cannot anticipate the beginning of the other’s utterance
ü Accidentally overlapped utterances may not be heard properly
Ø The trouble needs to be “repaired” somehow (Schegloff, 1977)
2. Not accidental overlap • Included in the Other
• Turn-‐taking with Overlap
• Simultaneous End
Overlapping utterances can be classified into five types (Figure 4), from a viewpoint of when the latter utterance starts and stops overlapping with the former3: (a) Simultaneous Start (two utterances are started simultaneously, and either of them is completed before the other), (b) Included in the Other (the latter is started after the former is started, and the latter is completed before the former is completed), (c) Turn-Taking with Overlap (the latter is started after the former is started, and the former is completed before the latter is completed), (d) Simultaneous End (the latter is started after the former is started, and the two utterances are completed simultaneously), (e) Simultaneous Start and End (two utterances are started and completed simultaneously) (modified after [7]).
3 In this paper, “simultaneously” means that the latter utterance is started less than 100ms after
the former utterance is started, and “One utterance is started after (completed before) the other” means that one utterance is started (completed) 100ms or more after (before) the other.
(a) Simultaneous Start F: utterance L: utterance (b) Included in the Other F: utterance L: utterance (c) Turn-Taking with Overlap F: utterance L: utterance
(d) Simultaneous End F: utterance L: utterance (e) Simultaneous Start and End F: utterance L: utterance * F: the former speaker, L: the latter speaker
Fig. 4. The five types of overlapping utterances.
Among the five types, type (a) and (e) has a different feature from the others. In general, when a hearer starts to overlap with the speaker’s utterance, the latter speaker, more or less, intends or expects to make his/her own utterance overlap with the former’s utterance. Therefore, the latter utterance never starts before the former starts. However, as for type (a) and (e), two utterances “accidentally” overlap, for neither of the two speakers can anticipate the beginning of the other’s utterance. When two participants start to speak at the same time, one of them or both of them may not be heard or understood completely. In such cases, the speaker him/herself or the others should start to “repair” the insufficiently understood utterances [8].
Nevertheless, in table cooking, it may not be frequent that accidental overlap of utterances (type (a) or (e)) is repaired either by the participant who made the trouble or by the other participant(s). In fact, in our experiments of conversations with monja, all the accidentally overlapped utterances were not repaired. Of all the overlaps, 3 examples in phase (1) and 4 examples in phase (3) were type (a). There were no examples of type (e) in phase (1) and (3). While all the examples of (a) in phase (1) were properly repaired, some in phase (3) were not repaired, which is likely to be one of the interesting aspects of interactions in table cooking.
Overlapping utterances can be classified into five types (Figure 4), from a viewpoint of when the latter utterance starts and stops overlapping with the former3: (a) Simultaneous Start (two utterances are started simultaneously, and either of them is completed before the other), (b) Included in the Other (the latter is started after the former is started, and the latter is completed before the former is completed), (c) Turn-Taking with Overlap (the latter is started after the former is started, and the former is completed before the latter is completed), (d) Simultaneous End (the latter is started after the former is started, and the two utterances are completed simultaneously), (e) Simultaneous Start and End (two utterances are started and completed simultaneously) (modified after [7]).
3 In this paper, “simultaneously” means that the latter utterance is started less than 100ms after
the former utterance is started, and “One utterance is started after (completed before) the other” means that one utterance is started (completed) 100ms or more after (before) the other.
(a) Simultaneous Start F: utterance L: utterance (b) Included in the Other F: utterance L: utterance (c) Turn-Taking with Overlap F: utterance L: utterance
(d) Simultaneous End F: utterance L: utterance (e) Simultaneous Start and End F: utterance L: utterance * F: the former speaker, L: the latter speaker
Fig. 4. The five types of overlapping utterances.
Among the five types, type (a) and (e) has a different feature from the others. In general, when a hearer starts to overlap with the speaker’s utterance, the latter speaker, more or less, intends or expects to make his/her own utterance overlap with the former’s utterance. Therefore, the latter utterance never starts before the former starts. However, as for type (a) and (e), two utterances “accidentally” overlap, for neither of the two speakers can anticipate the beginning of the other’s utterance. When two participants start to speak at the same time, one of them or both of them may not be heard or understood completely. In such cases, the speaker him/herself or the others should start to “repair” the insufficiently understood utterances [8].
Nevertheless, in table cooking, it may not be frequent that accidental overlap of utterances (type (a) or (e)) is repaired either by the participant who made the trouble or by the other participant(s). In fact, in our experiments of conversations with monja, all the accidentally overlapped utterances were not repaired. Of all the overlaps, 3 examples in phase (1) and 4 examples in phase (3) were type (a). There were no examples of type (e) in phase (1) and (3). While all the examples of (a) in phase (1) were properly repaired, some in phase (3) were not repaired, which is likely to be one of the interesting aspects of interactions in table cooking.
*F: the former speaker, L: the latter speaker
Viewpoint: when the latter utterance starts and stops overlapping with the former
Overlapping utterances can be classified into five types (Figure 4), from a viewpoint of when the latter utterance starts and stops overlapping with the former3: (a) Simultaneous Start (two utterances are started simultaneously, and either of them is completed before the other), (b) Included in the Other (the latter is started after the former is started, and the latter is completed before the former is completed), (c) Turn-Taking with Overlap (the latter is started after the former is started, and the former is completed before the latter is completed), (d) Simultaneous End (the latter is started after the former is started, and the two utterances are completed simultaneously), (e) Simultaneous Start and End (two utterances are started and completed simultaneously) (modified after [7]).
3 In this paper, “simultaneously” means that the latter utterance is started less than 100ms after
the former utterance is started, and “One utterance is started after (completed before) the other” means that one utterance is started (completed) 100ms or more after (before) the other.
(a) Simultaneous Start F: utterance L: utterance (b) Included in the Other F: utterance L: utterance (c) Turn-Taking with Overlap F: utterance L: utterance
(d) Simultaneous End F: utterance L: utterance (e) Simultaneous Start and End F: utterance L: utterance * F: the former speaker, L: the latter speaker
Fig. 4. The five types of overlapping utterances.
Among the five types, type (a) and (e) has a different feature from the others. In general, when a hearer starts to overlap with the speaker’s utterance, the latter speaker, more or less, intends or expects to make his/her own utterance overlap with the former’s utterance. Therefore, the latter utterance never starts before the former starts. However, as for type (a) and (e), two utterances “accidentally” overlap, for neither of the two speakers can anticipate the beginning of the other’s utterance. When two participants start to speak at the same time, one of them or both of them may not be heard or understood completely. In such cases, the speaker him/herself or the others should start to “repair” the insufficiently understood utterances [8].
Nevertheless, in table cooking, it may not be frequent that accidental overlap of utterances (type (a) or (e)) is repaired either by the participant who made the trouble or by the other participant(s). In fact, in our experiments of conversations with monja, all the accidentally overlapped utterances were not repaired. Of all the overlaps, 3 examples in phase (1) and 4 examples in phase (3) were type (a). There were no examples of type (e) in phase (1) and (3). While all the examples of (a) in phase (1) were properly repaired, some in phase (3) were not repaired, which is likely to be one of the interesting aspects of interactions in table cooking.
Overlapping utterances can be classified into five types (Figure 4), from a viewpoint of when the latter utterance starts and stops overlapping with the former3: (a) Simultaneous Start (two utterances are started simultaneously, and either of them is completed before the other), (b) Included in the Other (the latter is started after the former is started, and the latter is completed before the former is completed), (c) Turn-Taking with Overlap (the latter is started after the former is started, and the former is completed before the latter is completed), (d) Simultaneous End (the latter is started after the former is started, and the two utterances are completed simultaneously), (e) Simultaneous Start and End (two utterances are started and completed simultaneously) (modified after [7]).
3 In this paper, “simultaneously” means that the latter utterance is started less than 100ms after
the former utterance is started, and “One utterance is started after (completed before) the other” means that one utterance is started (completed) 100ms or more after (before) the other.
(a) Simultaneous Start F: utterance L: utterance (b) Included in the Other F: utterance L: utterance (c) Turn-Taking with Overlap F: utterance L: utterance
(d) Simultaneous End F: utterance L: utterance (e) Simultaneous Start and End F: utterance L: utterance * F: the former speaker, L: the latter speaker
Fig. 4. The five types of overlapping utterances.
Among the five types, type (a) and (e) has a different feature from the others. In general, when a hearer starts to overlap with the speaker’s utterance, the latter speaker, more or less, intends or expects to make his/her own utterance overlap with the former’s utterance. Therefore, the latter utterance never starts before the former starts. However, as for type (a) and (e), two utterances “accidentally” overlap, for neither of the two speakers can anticipate the beginning of the other’s utterance. When two participants start to speak at the same time, one of them or both of them may not be heard or understood completely. In such cases, the speaker him/herself or the others should start to “repair” the insufficiently understood utterances [8].
Nevertheless, in table cooking, it may not be frequent that accidental overlap of utterances (type (a) or (e)) is repaired either by the participant who made the trouble or by the other participant(s). In fact, in our experiments of conversations with monja, all the accidentally overlapped utterances were not repaired. Of all the overlaps, 3 examples in phase (1) and 4 examples in phase (3) were type (a). There were no examples of type (e) in phase (1) and (3). While all the examples of (a) in phase (1) were properly repaired, some in phase (3) were not repaired, which is likely to be one of the interesting aspects of interactions in table cooking.
Overlapping utterances can be classified into five types (Figure 4), from a viewpoint of when the latter utterance starts and stops overlapping with the former3: (a) Simultaneous Start (two utterances are started simultaneously, and either of them is completed before the other), (b) Included in the Other (the latter is started after the former is started, and the latter is completed before the former is completed), (c) Turn-Taking with Overlap (the latter is started after the former is started, and the former is completed before the latter is completed), (d) Simultaneous End (the latter is started after the former is started, and the two utterances are completed simultaneously), (e) Simultaneous Start and End (two utterances are started and completed simultaneously) (modified after [7]).
3 In this paper, “simultaneously” means that the latter utterance is started less than 100ms after
the former utterance is started, and “One utterance is started after (completed before) the other” means that one utterance is started (completed) 100ms or more after (before) the other.
(a) Simultaneous Start F: utterance L: utterance (b) Included in the Other F: utterance L: utterance (c) Turn-Taking with Overlap F: utterance L: utterance
(d) Simultaneous End F: utterance L: utterance (e) Simultaneous Start and End F: utterance L: utterance * F: the former speaker, L: the latter speaker
Fig. 4. The five types of overlapping utterances.
Among the five types, type (a) and (e) has a different feature from the others. In general, when a hearer starts to overlap with the speaker’s utterance, the latter speaker, more or less, intends or expects to make his/her own utterance overlap with the former’s utterance. Therefore, the latter utterance never starts before the former starts. However, as for type (a) and (e), two utterances “accidentally” overlap, for neither of the two speakers can anticipate the beginning of the other’s utterance. When two participants start to speak at the same time, one of them or both of them may not be heard or understood completely. In such cases, the speaker him/herself or the others should start to “repair” the insufficiently understood utterances [8].
Nevertheless, in table cooking, it may not be frequent that accidental overlap of utterances (type (a) or (e)) is repaired either by the participant who made the trouble or by the other participant(s). In fact, in our experiments of conversations with monja, all the accidentally overlapped utterances were not repaired. Of all the overlaps, 3 examples in phase (1) and 4 examples in phase (3) were type (a). There were no examples of type (e) in phase (1) and (3). While all the examples of (a) in phase (1) were properly repaired, some in phase (3) were not repaired, which is likely to be one of the interesting aspects of interactions in table cooking.
Accidental Overlaps Repaired Insufficiently • In table cooking, accidental overlaps may not be repaired
Table. Accidental overlaps (A.O.) in Phase (1) & (3) (times)
• Accidental overlaps were not so frequent ü However, when they occurred in cooking, many of them were not
repaired
Total of overlaps
Total of A.O.
Repaired A.O.
Not repaired A.O.
(1) Deciding what to eat 40 3 3 0
(3) Cooking monja-‐yaki 39 4 2 2
Transcript (1): Accidental Overlap Completely Repaired in Deciding What to Eat
• 03U and 04S were accidentally overlapped • Self-‐repair by 05U is smoothly accomplished ü 05U combined her previous question with that of 04S
• U repaired the overlap to make sure that H understands their questions
�ŋž� ��
�
�������Ɨkx���kxKl\JSl=Y�����kxKl^ǧ<AÍ@q[<xXL�� When I, I try to have monja, I don't recognize the difference of monjas. ��������p]���įðv�#�@�#�rgL���turning her eyes on participant S��� Do you understand? �����������>�Z=<=GY��������� Well, what does it mean? ������������¹�^�ǧ�<���� Difference of what? � ������ð^�ǧ<TWİK�� Is it the difference of the taste? �������������ÇǪkxKl��TWkxKl���\[s All monjas will be monjas. �
� �
← Accidental Overlap
← Self-‐repair
Transcript (2): Accidental Overlap not Repaired in Cooking
• 01U addressed S and asked a question to S • 02S answered the question by U • 04S and 05H started to speak simultaneously and overlapped ü 04S: to tell an additional answer to 01U ü 05H: to tell new information about the birthplace of monja-‐yaki Ø 05H was supposed to be answered or clarified by someone
�ŋž�����
�
� �������A������Kl[<����
� ����?�kSA]��
�� ��������kSA���
����^9���A������Kl[<:Y<=����\đJWȇ��_°ƯÁĤwLJ
hsȈJ@JLD\�kƳėÁĤJp=YJWȇ°ƯÁĤYƳėÁĤ^ǭƽAǕBs���
�
�ŋž�� ��
�
�������ƄƹIx_���>���kxKlTWdzƾ����dzś�����
�������dzś�Rp��
��������������;���������
����������dzƾ[�<�
� ��������ŕĔ����
�
�������ƄƹIx_���>���kxKlTWdzƾ����dzś������ � � � Rui*, well, are monjas from the Kansai ((region))? Or the Kanto ((region))? �������dzś�Rp��� � � � ((It is from)) the Kanto. ��������������;���������� � � � � � Oh. ����������dzƾ[�<�� � � � ((There are)) not ((any monjas)) in Kansai. � ��������ŕĔ����� � � � ((Is it from)) Tsukishima? �
�������ƄƹIx_���>���kxKlTWdzƾ����dzś������ � � � Rui*, well, are monjas from the Kansai ((region))? Or the Kanto ((region))? ��3!������������������������������������������������
�
�������dzś�Rp��� � � � ((It is from)) the Kanto. ��������������;���������� � � � � � Oh. ��3!�������������������
�
����������dzƾ[�<�� � � � ((There are)) not ((any monjas)) in Kansai. � ��������ŕĔ����� � � � ((Is it from)) Tsukishima? ��3!���������������
Accidental Overlap
((from then on, nobody answered 05H or repaired the trouble)) * Rui is participant S.
Gazing at Monja Hampers Smooth Turn-‐taking • Why didn’t S and U repair or answer 05H? ü S may not have heard 05H clearly ü U must have noticed H saying something Ø Just after 05H started to speak, U looked at H in a moment Ø Nevertheless, U did not respond to 05H
Monja-‐yaki
Monja-‐yaki
Monja-‐yaki
U
S
H
U Monja-‐yaki
S Monja H S Monja
Kanto [dayo.]
[Ah]h.
[[Kansai na]i.
[[Tsukishima?]
U
S
H
Gaz
e Utteran
ce
05H
* In this scene, only H is cooking.
Rui san wa (.) e (.) monja tte kansai? (.) Kanto?
Gazing at Monja Hampers Smooth Turn-‐taking • Why didn’t U answer 05H even though she noticed 05H? ü S and H concentrated on cooking, and U gave up talking with S and H? ü S did not gaze at U sufficiently even when taking with U Ø S turned his eyes on the monja before 01U completed the question Ø 02S was answered 01U with his eyes on the monja
ü U’s interest in the birthplace of monja may have been diminished • Typical of closure of topics in table cooking?
Monja-‐yaki
Monja-‐yaki
Monja-‐yaki
U
S
H
U Monja-‐yaki
S Monja H S Monja
Kanto [dayo.]
Rui san wa (.) e (.) monja tte kansai? (.) Kanto? [Ah]h.
[[Kansai na]i.
[[Tsukishima?]
U
S
H
Gaz
e Utteran
ce
05H
* In this scene, only H is cooking.
Transcript (3): Overlaps Accompanied with “Co-‐telling” (e.g. Toyama, 2012)
• S and U (more skilled) “co-‐told” H (less skilled) how to cook monja
• 01S tried to tell H to make a “dote” ü S was unable to vocalize “dote” quickly, and started a “word search” (e.g. Hayashi, 2003)
• 02U realized a collaborative instruction to H with overlap ü It was necessary to tell H how to go on cooking as soon as possible • 01S and 02U were allowed to overlap and repaired by no one
�ŋž�����
�
������������������Gt����;^������ùĵwºq�[<Y�� This, say... ((you)) have to make a “dote”*. ��������������������������ùĵºTW�¨ºTW��� Please make a “dote”*, (make a circle). ��3!��������������������������������������� ���������* A “dote” means a bank in Japanese. �
� �
Total Length and Average Length of Silences
• Why did silence occur more often while cooking?
(1) Deciding what to eat Ø Long time silence was not
allowed because of the urgent task of decision-‐making
(2) Waiting for the dishes Ø The participants had to focus on
talking without silence
(3) Cooking monja-‐yaki Ø Silence may have been allowed
because of all the participants focusing on cooking
0
0.5
1
1.5
2
2.5
0
20
40
60
80
(1) Deciding what to eat
(2) Waiting for the dishes
(3) Cooking monja-yaki
Length of silences among total length of each phase (%) Average length of silences (sec.)
Total Length and Average Length of Silences
• 1st author took part in and observed the conversation
• In cooking ü Even when long time silence
occurred, we did not necessarily feel embarrassed or unpleasant
ü Trustful relationships may have been produced?
• Further analysis in detail is needed 0
0.5
1
1.5
2
2.5
0
20
40
60
80
(1) Deciding what to eat
(2) Waiting for the dishes
(3) Cooking monja-yaki
Length of silences among total length of each phase (%) Average length of silences (sec.)
Situated Interaction in Cooking Acts • Overlaps and silence are generated by co-‐cooking ü Cooking acts permit us not to exchange glances
• Overlaps and silence are allowed because of the unique characteristics of monja-‐yaki
ü Instructing each other or confirming how to cook • instead of the most skilled participant monopolizing cooking
• Monja-‐yaki: interesting example of “co-‐creation” ü A goal can be achieved by all the 3 participants
References Toyama, E., Den, Y., and Kowaki, T.: How do we coordinate eating and speaking? (in Japanese). In Proceedings of the 65th Conference of Special Interest Group on Spoken Language Understanding and Dialogue Processing (SIG-‐SLUD) (2012) Enomoto, M. and Den, Y.: Will the Participant Gazed at by the Current Speaker be the Next Speaker? (in Japanese). The Japanese Journal of Language in Society (Japanese edition), Vol. 14, No. 1 (2011) 97-‐109 Toyama, E., Kikuchi, K., and Bono, M.: Joint Construction of Narrative Space: Coordination of gesture and sequence in Japanese three-‐party conversation. In Proceedings of International Workshop on Multimodality in Multispace Interaction (MiMI 2011) (2011) Sacks, H., Schegloff, E., and Jefferson, G.: A simplest systematics for the organization of turn-‐taking in conversation. Language, Vol. 50, No. 4 (1974) 696-‐735 Mukawa, M., Tokunaga, H., Yuasa, M., Tsuda, Y., Tateyama, K., and Kasamatsu, C.: Analysis on utterance behaviors embedded in eating actions: How are conversations and hand-‐mouth-‐motions controlled in three-‐party table talk? (in Japanese). The IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences (Japanese edition), Vol. J94-‐A, No.7 (2011) 500-‐508 Nishizaka, A., Kushida, S., and Kumagai, T.: Introduction (in Japanese). Special Issue: Language Use in Interaction, The Japanese Journal of Language in Society (Japanese edition), Vol. 10, No. 2 (2008) 13-‐15 Enomoto, M.: When does the hearer start his turn? The turn-‐taking rules in Japanese conversation apply retrospectively after a possible completion point has passed (in Japanese). Cognitive Studies, Vol. 10, No. 2 (2003) 291-‐303 Schegloff, E.A., Jefferson, G. and Sacks, H.: The preference for self-‐correction in the organization of repair in conversation, Language, Vol. 53, No. 2 (1977) 361–382 Hayashi, M.: Language and the Body as Resources for Collaborative Action: A Study of Word Searches in Japanese Conversation. Research on Language and Social Interaction, Vol. 36, No. 2 (2003) 109-‐141