File size: 82,477 Bytes
6fa4bc9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 | {
"paper_id": "P01-1016",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T09:29:53.351073Z"
},
"title": "Non-Verbal Cues for Discourse Structure",
"authors": [
{
"first": "Justine",
"middle": [],
"last": "Cassell",
"suffix": "",
"affiliation": {},
"email": "justine@media.mit.edu"
},
{
"first": "Yukiko",
"middle": [
"I"
],
"last": "Nakano",
"suffix": "",
"affiliation": {},
"email": "yukiko@media.mit.edu"
},
{
"first": "Timothy",
"middle": [
"W"
],
"last": "Bickmore",
"suffix": "",
"affiliation": {},
"email": "bickmore@media.mit.edu"
},
{
"first": "Candace",
"middle": [
"L"
],
"last": "Sidner",
"suffix": "",
"affiliation": {},
"email": "sidner@merl.com"
},
{
"first": "Charles",
"middle": [],
"last": "Rich",
"suffix": "",
"affiliation": {},
"email": "rich@merl.com"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "This paper addresses the issue of designing embodied conversational agents that exhibit appropriate posture shifts during dialogues with human users. Previous research has noted the importance of hand gestures, eye gaze and head nods in conversations between embodied agents and humans. We present an analysis of human monologues and dialogues that suggests that postural shifts can be predicted as a function of discourse state in monologues, and discourse and conversation state in dialogues. On the basis of these findings, we have implemented an embodied conversational agent that uses Collagen in such a way as to generate postural shifts.",
"pdf_parse": {
"paper_id": "P01-1016",
"_pdf_hash": "",
"abstract": [
{
"text": "This paper addresses the issue of designing embodied conversational agents that exhibit appropriate posture shifts during dialogues with human users. Previous research has noted the importance of hand gestures, eye gaze and head nods in conversations between embodied agents and humans. We present an analysis of human monologues and dialogues that suggests that postural shifts can be predicted as a function of discourse state in monologues, and discourse and conversation state in dialogues. On the basis of these findings, we have implemented an embodied conversational agent that uses Collagen in such a way as to generate postural shifts.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "This paper provides empirical support for the relationship between posture shifts and discourse structure, and then derives an algorithm for generating posture shifts in an animated embodied conversational agent from discourse states produced by the middleware architecture known as Collagen [18] . Other nonverbal behaviors have been shown to be correlated with the underlying conversational structure and information structure of discourse. For example, gaze shifts towards the listener correlate with a shift in conversational turn (from the conversational participants' perspective, they can be seen as a signal that the floor is available). Gestures correlate with rhematic content in accompanying language (from the conversational participants' perspective, these behaviors can be seen as a signal that accompanying speech is of high interest). A better understanding of the role of nonverbal behaviors in conveying discourse structures enables improvements in the naturalness of embodied dialogue systems, such as embodied conversational agents, as well as contributing to algorithms for recognizing discourse structure in speech-understanding systems.",
"cite_spans": [
{
"start": 292,
"end": 296,
"text": "[18]",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "Previous work, however, has not addressed major body shifts during discourse, nor has it addressed the nonverbal correlates of topic shifts.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "Only recently have computational linguists begun to examine the association of nonverbal behaviors and language. In this section we review research by non-computational linguists and discuss how this research has been employed to formulate algorithms for natural language generation or understanding. About three-quarters of all clauses in descriptive discourse are accompanied by gestures [17] , and within those clauses, the most effortful part of gestures tends to co-occur with or just before the phonologically most prominent syllable of the accompanying speech [13] . It has been shown that when speech is ambiguous or in a speech situation with some noise, listeners rely on gestural cues [22] (and, the higher the noise-tosignal ratio, the more facilitation by gesture). Even when gestural content overlaps with speech (reported to be the case in roughly 50% of utterances, for descriptive discourse), gesture often emphasizes information that is also focused pragmatically by mechanisms like prosody in speech. In fact, the semantic and pragmatic compatibility in the gesture-speech relationship recalls the interaction of words and graphics in multimodal presentations [11] .",
"cite_spans": [
{
"start": 390,
"end": 394,
"text": "[17]",
"ref_id": "BIBREF17"
},
{
"start": 567,
"end": 571,
"text": "[13]",
"ref_id": "BIBREF13"
},
{
"start": 696,
"end": 700,
"text": "[22]",
"ref_id": "BIBREF23"
},
{
"start": 1179,
"end": 1183,
"text": "[11]",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2."
},
{
"text": "On the basis of results such as these, several researchers have built animated embodied conversational agents that ally synthesized speech with animated hand gestures. For example, Lester et al. [15] generate deictic gestures and choose referring expressions as a function of the potential ambiguity and proximity of objects referred to. Rickel and Johnson [19] 's pedagogical agent produces a deictic gesture at the beginning of explanations about objects. Andr\u00e9 et al. [1] generate pointing gestures as a sub-action of the rhetorical action of labeling, in turn a sub-action of elaborating. Cassell and Stone [3] generate either speech, gesture, or a combination of the two, as a function of the information structure status and surprise value of the discourse entity.",
"cite_spans": [
{
"start": 195,
"end": 199,
"text": "[15]",
"ref_id": "BIBREF15"
},
{
"start": 357,
"end": 361,
"text": "[19]",
"ref_id": "BIBREF19"
},
{
"start": 471,
"end": 474,
"text": "[1]",
"ref_id": "BIBREF0"
},
{
"start": 611,
"end": 614,
"text": "[3]",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2."
},
{
"text": "Head and eye movement has also been examined in the context of discourse and conversation. Looking away from one's interlocutor has been correlated with the beginning of turns. From the speaker's point of view, this look away may prevent an overload of visual and linguistic information. On the other hand, during the execution phase of an utterance, speakers look more often at listeners. Head nods and eyebrow raises are correlated with emphasized linguistic items -such as words accompanied by pitch accents [7] .",
"cite_spans": [
{
"start": 511,
"end": 514,
"text": "[7]",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2."
},
{
"text": "Some eye movements occur primarily at the ends of utterances and at grammatical boundaries, and appear to function as synchronization signals. That is, one may request a response from a listener by looking at the listener, and suppress the listener's response by looking away. Likewise, in order to offer the floor, a speaker may gaze at the listener at the end of the utterance. When the listener wants the floor, s/he may look at and slightly up at the speaker [10] . It should be noted that turn taking only partially accounts for eye gaze behavior in discourse. A better explanation for gaze behavior integrates turn taking with the information structure of the propositional content of an utterance [5] . Specifically, the beginning of themes are frequently accompanied by a look-away from the hearer, and the beginning of rhemes are frequently accompanied by a look-toward the hearer. When these categories are co-temporaneous with turn construction, then they are strongly predictive of gaze behavior.",
"cite_spans": [
{
"start": 463,
"end": 467,
"text": "[10]",
"ref_id": "BIBREF10"
},
{
"start": 704,
"end": 707,
"text": "[5]",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2."
},
{
"text": "Results such as these have led researchers to generate eye gaze and head movements in animated embodied conversational agents. Takeuchi and Nagao, for example, [21] generate gaze and head nod behaviors in a \"talking head.\" Cassell et al. [2] generate eye gaze and head nods as a function of turn taking behavior, head turns just before an utterance, and eyebrow raises as a function of emphasis.",
"cite_spans": [
{
"start": 160,
"end": 164,
"text": "[21]",
"ref_id": "BIBREF22"
},
{
"start": 238,
"end": 241,
"text": "[2]",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2."
},
{
"text": "To our knowledge, research on posture shifts and other gross body movements, has not been used in the design or implementation of computational systems. In fact, although a number of conversational analysts and ethnomethodologists have described posture shifts in conversation, their studies have been qualitative in nature, and difficult to reformulate as the basis of algorithms for the generation of language and posture. Nevertheless, researchers in the non-computational fields have discussed posture shifts extensively. Kendon [13] reports a hierarchy in the organization of movement such that the smaller limbs such as the fingers and hands engage in more frequent movements, while the trunk and lower limbs change relatively rarely.",
"cite_spans": [
{
"start": 533,
"end": 537,
"text": "[13]",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2."
},
{
"text": "A number of researchers have noted that changes in physical distance during interaction seem to accompany changes in the topic or in the social relationship between speakers. For example Condon and Osgton [9] have suggested that in a speaking individual the changes in these more slowly changing body parts occur at the boundaries of the larger units in the flow of speech.",
"cite_spans": [
{
"start": 205,
"end": 208,
"text": "[9]",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2."
},
{
"text": "Scheflen (1973) also reports that posture shifts and other general body movements appear to mark the points of change between one major unit of communicative activity and another. Blom & Gumperz (1972) identify posture changes and changes in the spatial relationship between two speakers as indicators of what they term \"situational shifts\" --momentary changes in the mutual rights and obligations between speakers accompanied by shifts in language style. Erickson (1975) concludes that proxemic shifts seem to be markers of 'important' segments. In his analysis of college counseling interviews, they occurred more frequently than any other coded indicator of segment changes, and were therefore the best predictor of new segments in the data. Unfortunately, in none of these studies are statistics provided, and their analyses rely on intuitive definitions of discourse segment or \"major shift\". For this reason, we carried out our own empirical study.",
"cite_spans": [
{
"start": 180,
"end": 201,
"text": "Blom & Gumperz (1972)",
"ref_id": null
},
{
"start": 456,
"end": 471,
"text": "Erickson (1975)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2."
},
{
"text": "Videotaped \"pseudo-monologues\" and dialogues were used as the basis for the current study. In \"pseudo-monologues,\" subjects were asked to describe each of the rooms in their home, then give directions between four pairs of locations they knew well (e.g., home and the grocery store). The experimenter acted as a listener, only providing backchannel feedback (head nods, smiles and paraverbals such as \"uh-huh\"). For dialogues, two subjects were asked to generate an idea for a class project that they would both like to work on, including: 1) what they would work on; 2) where they would work on it (including facilities, etc.), and 3) when they would work on it. Subjects stood in both conditions and were told to perform their tasks in 5-10 minutes.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Empirical Study",
"sec_num": "3."
},
{
"text": "The pseudo-monologue condition (pseudo-because there was in fact an interlocutor, although he gave backchannel feedback only and never took the turn) allowed us to investigate the relationship between discourse structure and posture shift independent of turn structure. The two tasks were constructed to allow us to identify exactly where discourse segment boundaries would be placed.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Empirical Study",
"sec_num": "3."
},
{
"text": "The video data was transcribed and coded for three features: discourse segment boundaries, turn boundaries, and posture shifts. A discourse segment is taken to be an aggregation of utterances and sub-segments that convey the discourse segment purpose, which is an intention that leads to the segment initiation [12] . In this study we chose initially to look at high-level discourse segmentation phenomena rather than those discourse segments embedded deeper in the discourse. Thus, the time points at which the assigned task topics were started served as segmentation points. Turn boundaries were coded (for dialogues only) as the point in time in which the start or end of an utterance cooccurred with a change in speaker, but excluding backchannel feedback. Turn overlaps were coded as open-floor time. We defined a posture shift as a motion or a position shift for a part of the human body, excluding hands and eyes (which we have dealt with in other work). Posture shifts were coded with start and end time of occurrence (duration), body part in play (for this paper we divided the body at the waistline and compared upper body vs. lower body shifts), and an estimated energy level of the posture shift. Energy level was normalized for each subject by taking the largest posture shift observed for each subject as 100% and coding all other posture shift energies relative to the 100% case. Posture shifts that occurred as part of gesture or were clearly intentionally generated (e.g., turning one's body while giving directions) were not coded.",
"cite_spans": [
{
"start": 311,
"end": 315,
"text": "[12]",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Empirical Study",
"sec_num": "3."
},
{
"text": "Data from seven monologues and five dialogues were transcribed, and then coded and analyzed independently by two raters. A total of 70.5 minutes of data was analyzed (42.5 minutes of dialogue and 29.2 minutes of monologue). A total of 67 discourse segments were identified (25 in the dialogues and 42 in the monologues), which constituted 407 turns in the dialogue data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "4."
},
{
"text": "We used the instructions given to subjects concerning the topics to discuss as segmentation boundaries. In future research, we will address the smaller discourse segmentation. For posture shift coding, raters coded all posture shifts independently, and then calculated reliability on the transcripts of one monologue (5.2 minutes) and both speakers from one dialogue (8.5 minutes). Agreement on the presence of an upper body or lower body posture shift in a particular location (taking location to be a 1second window that contains all of or a part of a posture shift) for these three speakers was 89% (kappa = .64). For interrater reliability of the coding of energy level, a Spearman's rho revealed a correlation coefficient of .48 (p<.01).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "4."
},
{
"text": "Posture shifts occurred regularly throughout the data (an average of 15 per speaker in both pseudo-monologues and dialogues). This, together with the fact that the majority of time was spent within discourse segments and within turns (rather than between segments), led us to normalize our posture shift data for comparison purposes. For relatively brief intervals (interdiscourse-segment and inter-turn) normalization by number of inter-segment occurrences was sufficient (ps/int), however, for long intervals (intra-discourse segment and intra-turn) we needed to normalize by time to obtain meaningful comparisons. For this normalization metric we looked at posture-shifts-per-second (ps/s). This gave us a mean average of .06 posture shifts/second (ps/s) in the monologues (SD=.07), and .07 posture shifts/second in the dialogues (SD=.08). Our initial analysis compared posture shifts made by the current speaker within discourse segments (intra-dseg) to those produced at the boundaries of discourse segments (inter-dseg). It can be seen (in Table 4 .1.1) that posture shifts occur an order of magnitude more frequently at discourse segment boundaries than within discourse segments in both monologues and dialogues. Posture shifts also tend to be more energetic at discourse segment boundaries (F(1,251)=10.4; p<0.001). Initially, we classified data as being inter-or intra-turn. Table 4 .1.2 shows that turn structure does have an influence on posture shifts; subjects were five times more likely to exhibit a shift at a boundary than within a turn. An interaction exists between turns and discourse segments such that discourse segment boundaries are ten times more likely to co-occur with turn changes than within turns. Both turn and discourse structure exhibit an influence on posture shifts, with discourse having the most predictive value. Starting a turn while starting a new discourse segment is marked with a posture shift roughly 10 times more often than when starting a turn while staying within discourse segment. We noticed, however, that posture shifts appeared to congregate at the beginnings or ends of turn boundaries, and so our subsequent analyses examined start-turns, midturns and end-turns. It is clear from these results that posture is indeed correlated with discourse state, such that speakers generate a posture shift when initiating a new discourse segment, which is often at the boundary between turns.",
"cite_spans": [],
"ref_spans": [
{
"start": 1046,
"end": 1053,
"text": "Table 4",
"ref_id": "TABREF0"
},
{
"start": 1385,
"end": 1392,
"text": "Table 4",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "Analysis",
"sec_num": "4.1"
},
{
"text": "In addition to looking at the occurrence and energy of posture shifts we also analyzed the distributions of upper vs. lower body shifts and the duration of posture shifts. Speaker upper body shifts were found to be used more frequently at the start of turns (48%) than at the middle of turns (36%) or end of turns (18%) ( dependence on discourse structure. Finally, speaker posture shift duration was found to change significantly as a function of both turn and discourse structure (see Figure 4.1.3) . At the start of turns, posture shift duration is approximately the same whether a new topic is introduced or not (2.5 seconds). However, when ending a turn, speakers move significantly longer (7.0 seconds) when finishing a topic than when the topic is continued by the other interlocutor (2.7 seconds) (F(1,148)=17.9; p<0.001). ",
"cite_spans": [],
"ref_spans": [
{
"start": 487,
"end": 500,
"text": "Figure 4.1.3)",
"ref_id": null
}
],
"eq_spans": [],
"section": "Analysis",
"sec_num": "4.1"
},
{
"text": "In the following sections we discuss how the results of the empirical study were integrated along with Collagen into our existent embodied conversational agent, Rea.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "System",
"sec_num": "5."
},
{
"text": "Rea is an embodied conversational agent that interacts with a user in the real estate agent domain [2] . The system architecture of Rea is shown in Figure 5 .1. Rea takes input from a microphone and two cameras in order to sense the user's speech and gesture. The UM interprets and integrates this multimodal input and outputs a unified semantic representation. The Understanding Module then sends the output to Collagen as the Dialogue Manager.",
"cite_spans": [
{
"start": 99,
"end": 102,
"text": "[2]",
"ref_id": "BIBREF1"
}
],
"ref_spans": [
{
"start": 148,
"end": 156,
"text": "Figure 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "System Architecture",
"sec_num": "5.1"
},
{
"text": "Collagen, as further discussed below, maintains the state of the dialogue as shared between Rea and a user. The Reaction Module decides Rea's next action based on the discourse state maintained by Collagen. It also assigns information structure to output utterances so that gestures can be appropriately generated. The semantic representation of the action, including verbal and non-verbal behaviors, is sent to the Generation Module which generates surface linguistic expressions and gestures, including a set of instructions to achieve synchronization between animation and speech. These instructions are executed by a 3D animation renderer and a text-to-speech system. Table 5 .1 shows the associations between discourse and conversational state that Rea is currently able to handle. In other work we have discussed how Rea deals with the association between information structure and gesture [6] . In the following sections, we focus on Rea's generation of posture shifts. ",
"cite_spans": [
{
"start": 896,
"end": 899,
"text": "[6]",
"ref_id": "BIBREF5"
}
],
"ref_spans": [
{
"start": 672,
"end": 679,
"text": "Table 5",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "System Architecture",
"sec_num": "5.1"
},
{
"text": "Collagen TM is JAVA middleware for building COLLAborative interface AGENts to work with users on interface applications. Collagen is designed with the capability to participate in collaboration and conversation, based on [12] , [16] . Collagen updates the focus stack and recipe tree using a combination of the discourse interpretation algorithm of [16] and plan recognition algorithms of [14] . It takes as input user and system utterances and interface actions, and accesses a library of recipes describing actions in the domain. After updating the discourse state, Collagen makes three resources available to the interface agent: focus of attention (using the focus stack), segmented interaction history (of completed segments) and an agenda of next possible actions created from the focus stack and recipe tree.",
"cite_spans": [
{
"start": 221,
"end": 225,
"text": "[12]",
"ref_id": "BIBREF12"
},
{
"start": 228,
"end": 232,
"text": "[16]",
"ref_id": "BIBREF16"
},
{
"start": 349,
"end": 353,
"text": "[16]",
"ref_id": "BIBREF16"
},
{
"start": 389,
"end": 393,
"text": "[14]",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The Collagen dialogue manager",
"sec_num": "5.2"
},
{
"text": "The Reaction Module works as a content planner in the Rea architecture, and also plays the role of an interface agent in Collagen. It has access to the discourse state and the agenda using APIs provided by Collagen. Based on the results reported above, we describe here how Rea plans her next nonverbal actions using the resources that Collagen maintains.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Generation",
"sec_num": "5.3"
},
{
"text": "The empirical study revealed that posture shifts are distributed with respect to discourse segment and turn boundaries, and that the form of a posture shift differs according to these codeterminants. Therefore, generation of posture shifts in Rea is determined according to these two factors, with Collagen contributing information about current discourse state.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Generation",
"sec_num": "5.3"
},
{
"text": "Any posture shift that occurs between the end of one discourse segment and the beginning of the next is defined as an inter-discourse segment posture shift. In order to elaborate different generation rules for inter-vs. intra-discourse segments, Rea judges (D1) whether the next utterance starts a new topic, or contributes to the current discourse purpose, (D2) whether the next utterance is expected to finish a segment.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discourse structure information",
"sec_num": "5.3.1"
},
{
"text": "First, (D1) is calculated by referring to the focus stack and agenda. In planning a next action, Rea accesses the goal agenda in Collagen and gets the content of her next utterance. She also accesses the focus stack and gets the current discourse purpose that is shared between her and the user. By comparing the current purpose and the purpose of her next utterance, Rea can judge whether the her next utterance contributes to the current discourse purpose or not. For example, if the current discourse purpose is to find a house to show the user (FindHouse), and the next utterance that Rea plans to say is as follows, then this utterance does not directly contribute to the current discourse purpose because it does not ask a parameter of FindHouse, and it introduces a new discourse purpose ShowHouse. In this case, Rea judges that there is a discourse segment boundary between the previous utterance and the next one (D1 = topic change).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discourse structure information",
"sec_num": "5.3.1"
},
{
"text": "In order to calculate (D2), Rea looks at the plan tree in Collagen, and judges whether the next utterance addresses the last goal in the current discourse purpose. If it is the case, Rea expects to finish the current discourse segment by the next utterance (D1 = finish topic). As for conversational structure, Rea needs to know; (T1) whether Rea is taking a new turn with the next utterance, or keeping her current turn for the next utterance, (T2) whether Rea's next utterance requires that the user respond.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discourse structure information",
"sec_num": "5.3.1"
},
{
"text": "First, (T1) is judged by referring to the dialogue history 1 . The dialogue history stores both system utterances and user utterances that occurred in the dialogue. In the history, each utterance is stored as a logical form based on an artificial discourse language [20] . As shown above in utterance (1), the first argument of the action indicates the speaker of the utterance; in this example, it is \"agent\". The turn boundary can be estimated by comparing the speaker of the previous utterance with the speaker of the next utterance. If the speaker of the previous utterance is not Rea, there is a turn boundary before the next utterance (T1 = take turn). If the speaker of the previous utterance is Rea, that means that Rea will keep the same turn for the next utterance (T1 = keep turn).",
"cite_spans": [
{
"start": 59,
"end": 60,
"text": "1",
"ref_id": "BIBREF0"
},
{
"start": 266,
"end": 270,
"text": "[20]",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discourse structure information",
"sec_num": "5.3.1"
},
{
"text": "Second, (T2) is judged by looking at the type of Rea's next utterance. For example, when Rea asks a question, as in utterance (1), Rea expects the user to answer the question. In this case, Rea must convey to the user that the system gives up the turn (T2 = give up turn).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discourse structure information",
"sec_num": "5.3.1"
},
{
"text": "Combining information about discourse structure (D1, D2) and conversation structure (T1, T2), the system decides on posture shifts 1 We currently maintain a dialogue history in Rea even though Collagen has one as well. This is in order to store and manipulate the information to generate hand gestures and assign intonational accents. This information will be integrated into Collagen in the near future.",
"cite_spans": [
{
"start": 131,
"end": 132,
"text": "1",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Deciding and selecting a posture shift",
"sec_num": "5.3.2"
},
{
"text": "for the beginning of the utterance and the end of the utterance. Rea decides to do or not to do a posture shift by calling a probabilistic function that looks up the probabilities in Table 5 .3.1.",
"cite_spans": [],
"ref_spans": [
{
"start": 183,
"end": 190,
"text": "Table 5",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "Deciding and selecting a posture shift",
"sec_num": "5.3.2"
},
{
"text": "A posture shift for the beginning of the utterance is decided based on the combination of (D1) and (T1). For example, if the combined factors match Case (a), the system decides to generate a posture shift with 54% probability for the beginning of the utterance. Note that in Case (d), that is, Rea keeps the turn without changing a topic, we cannot calculate a per interval posture shift rate. Instead, we use a posture shift rate normalized for time. This rate is used in the GenerationModule, which calculates the utterance duration and generates a posture shift during the utterance based on this posture shift rate. On the other hand, ending posture shifts are decided based on the combination of (D2) and (T2).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Deciding and selecting a posture shift",
"sec_num": "5.3.2"
},
{
"text": "For example, if the combined factors match Case (e), the system decides to generate a posture shift with 0.04% probability for the ending of the utterance. When Rea does decide to activate a posture shift, she then needs to choose which posture shift to perform. Our empirical data indicates that the energy level of the posture shift differs depending on whether there is a discourse segment boundary or not. Moreover the duration of a posture shift differs depending on the place in a turn: start-, mid-, or end-turn. Based on these results, we define posture shift selection rules for energy, duration, and body part. The correspondence with discourse information is shown in ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Deciding and selecting a posture shift",
"sec_num": "5.3.2"
},
{
"text": "continue 0.04/sec low Posture shifts for pseudo-monologues can be decided using the same mechanism as that for dialogue, but omitting conversation structure information.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "D1",
"sec_num": null
},
{
"text": "The probabilities are given in table Table 5 .3.2. For example, if Rea changes the topic with her next utterance, a posture shift is generated 84% of the time with high-energy motion. In other cases, the system randomly generates low-energy posture shifts 0.04 times per second. This case is judged to be a turn boundary within a discourse segment (Case (c)), and Rea decides to generate a posture shift at the beginning of the utterance with 13% probability. If Rea decides to shift posture she selects a low energy posture shift using either upper or lower body. In addition to a posture shift at the beginning of the utterance, Rea may also choose to generate a posture shift to end the turn. As utterance (2) expects the user to take the turn, and continue to work on the same discourse purpose, this is Case (f). Thus, the system generates an end utterance posture shift 11% of the time. If generated, a low energy posture shift is chosen. If a beginning and/or ending posture shifts are generated, they are sent to the GM, which calculates the schedule of these multimodal events and generates them. As a subdialogue of showing a house, in a discourse purpose [goal : DiscussFeature], Rea keeps the turn and continues to describe the house. We handle this type of interaction as a pseudo-monologue. Therefore, we can use table Table 5 .3.2 for deciding on posture shifts here. In utterance (27), Rea starts the discussion about the house, and takes the initiative. This is judged as Case (g), and a high energy body motion is generated 84% of the time. ",
"cite_spans": [],
"ref_spans": [
{
"start": 31,
"end": 44,
"text": "table Table 5",
"ref_id": "TABREF4"
},
{
"start": 1333,
"end": 1340,
"text": "Table 5",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "D1",
"sec_num": null
},
{
"text": "We have demonstrated a clear relationship between nonverbal behavior and discourse state, and shown how this finding can be incorporated into the generation of language and nonverbal behaviors for an embodied conversational agent.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion and Further work",
"sec_num": "7."
},
{
"text": "Speakers produce posture shifts at 53% of discourse segment boundaries, more frequently than they produce those shifts discourse segment-internally, and with more motion energy. Furthermore, there is a relationship between discourse structure and conversational structure such that when speakers initiate a new segment at the same time as starting a turn (the most frequent case by far), they are more likely to produce a posture shift; while when they end a discourse segment and a turn at the same time, their posture shifts last longer than when these categories do not co-occur.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion and Further work",
"sec_num": "7."
},
{
"text": "Although this paper reports results from a limited number of monologues and dialogues, the findings are promising. In addition, they point the way to a number of future directions, both within the study of posture and discourse, and more generally within the study of nonverbal behaviors in computational linguistics. First, given the relationship between conversational and information structure in [5] , a natural next step is to examine the three-way relationship between discourse state, conversational structure (turns), and information structure (theme/rheme). For the moment, we have demonstrated that posture shifts may signal boundaries of units; do they also signal the information content of units? Next, we need to look at finer segmentations of the discourse, to see whether larger and smaller discourse segments are distinguished through non-verbal means. Third, the question of listener posture is an important one. We found that a number of posture shifts were produced by the participant who was not speaking. More than half of these shifts were produced at the same time as a speaker shift, suggesting a kind of mirroring. In order to interpret these data, however, a more sensitive notion of turn structure is required, as one must be ready to define when exactly speakers and listeners shift roles. Also, of course, evaluation of the importance of such nonverbal behaviors to user interaction is essential. In a user study of our earlier Gandalf system [4] , users rated the agent's language skills significantly higher under test conditions in which Gandalf deployed conversational behaviors (gaze, head movement and limited gesture) than when these behaviors were disabled. Such an evaluation is also necessary for the Rea-posture system.",
"cite_spans": [
{
"start": 400,
"end": 403,
"text": "[5]",
"ref_id": "BIBREF4"
},
{
"start": 1473,
"end": 1476,
"text": "[4]",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion and Further work",
"sec_num": "7."
},
{
"text": "But, more generally, we need to test whether generating posture shifts of this sort actually serves as a signal to listeners, for example to initiative structure in task and dialogue [8] . These evaluations form part of our future research plans.",
"cite_spans": [
{
"start": 183,
"end": 186,
"text": "[8]",
"ref_id": "BIBREF7"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion and Further work",
"sec_num": "7."
}
],
"back_matter": [
{
"text": "This research was supported by MERL, France Telecom, AT&T, and the other generous sponsors of the MIT Media Lab. Thanks to the other members of the Gesture and Narrative Language Group, in particular Ian Gouldstone and Hannes Vilhj\u00e1lmsson.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "acknowledgement",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Employing AI methods to control the behavior of animated interface agents",
"authors": [
{
"first": "E",
"middle": [],
"last": "Andre",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Rist",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Muller",
"suffix": ""
}
],
"year": 1999,
"venue": "Applied Artificial Intelligence",
"volume": "13",
"issue": "",
"pages": "415--448",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Andre, E., Rist, T., & Muller, J., Employing AI methods to control the behavior of animated interface agents, Applied Artificial Intelligence, vol. 13, pp. 415-448, 1999.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Embodiment in Conversational Interfaces: Rea, Proc. of CHI 99",
"authors": [
{
"first": "J",
"middle": [],
"last": "Cassell",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Bickmore",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Billinghurst",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Campbell",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Vilhjalmsson",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Yan",
"suffix": ""
}
],
"year": 1999,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cassell, J., Bickmore, T., Billinghurst, M., Campbell, L., Chang, K., Vilhjalmsson, H., & Yan, H., Embodiment in Conversational Interfaces: Rea, Proc. of CHI 99, Pittsburgh, PA, ACM, 1999.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Coordination and context-dependence in the generation of embodied conversation",
"authors": [
{
"first": "J",
"middle": [],
"last": "Cassell",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Stone",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Yan",
"suffix": ""
}
],
"year": 2000,
"venue": "Proc. INLG",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cassell, J., Stone, M., & Yan, H., Coordination and context-dependence in the generation of embodied conversation, Proc. INLG 2000, Mitzpe Ramon, Israel, 2000.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "The Power of a Nod and a Glance: Envelope vs. Emotional Feedback in Animated Conversational Agents",
"authors": [
{
"first": "J",
"middle": [],
"last": "Cassell",
"suffix": ""
},
{
"first": "K",
"middle": [
"R"
],
"last": "Thorisson",
"suffix": ""
}
],
"year": 1999,
"venue": "Applied Art. Intell",
"volume": "13",
"issue": "",
"pages": "519--538",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cassell, J. and Thorisson, K. R., The Power of a Nod and a Glance: Envelope vs. Emotional Feedback in Animated Conversational Agents, Applied Art. Intell., vol. 13, pp. 519-538, 1999.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Discourse Structure: How Best to Model Multimodal Conversation",
"authors": [
{
"first": "J",
"middle": [],
"last": "Cassell",
"suffix": ""
},
{
"first": "O",
"middle": [],
"last": "Torres",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Prevost",
"suffix": ""
}
],
"year": 1999,
"venue": "Machine Conversations",
"volume": "",
"issue": "",
"pages": "143--154",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cassell, J., Torres, O., & Prevost, S., Turn Taking vs. Discourse Structure: How Best to Model Multimodal Conversation., in Machine Conversations, Y. Wilks, Ed. The Hague: Kluwer, 1999, pp. 143-154.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "BEAT: The Behavior Expression Animation Toolkit",
"authors": [
{
"first": "J",
"middle": [],
"last": "Cassell",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Vilhj\u00e1lmsson",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Bickmore",
"suffix": ""
}
],
"year": 2001,
"venue": "Proc. of SIGGRAPH",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cassell, J., Vilhj\u00e1lmsson, H., & Bickmore, T., BEAT: The Behavior Expression Animation Toolkit, Proc. of SIGGRAPH, ACM Press, 2001.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Discourse-Oriented Facial Displays in Conversation",
"authors": [
{
"first": "N",
"middle": [],
"last": "Chovil",
"suffix": ""
}
],
"year": 1992,
"venue": "Research on Language and Social Interaction",
"volume": "25",
"issue": "",
"pages": "163--194",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Chovil, N., Discourse-Oriented Facial Displays in Conversation, Research on Language and Social Interaction, vol. 25, pp. 163-194, 1992.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Initiative in Collaborative Interactions -Its Cues and Effects",
"authors": [
{
"first": "J",
"middle": [],
"last": "Chu-Carroll",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Brown",
"suffix": ""
}
],
"year": 1997,
"venue": "Proc. of AAAI Spring 1997 Symp. on Computational Models of Mixed Initiative",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Chu-Carroll, J. & Brown, M., Initiative in Collaborative Interactions -Its Cues and Effects, Proc. of AAAI Spring 1997 Symp. on Computational Models of Mixed Initiative, 1997.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Speech and body motion synchrony of the speaker-hearer, in The perception of language",
"authors": [
{
"first": "W",
"middle": [
"S"
],
"last": "Condon",
"suffix": ""
},
{
"first": "W",
"middle": [
"D"
],
"last": "Osgton",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Condon, W. S. & Osgton, W. D., Speech and body motion synchrony of the speaker-hearer, in The perception of language, D. Horton & J.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "On the structure of speaker-auditor interaction during speaking turns",
"authors": [
{
"first": "S",
"middle": [],
"last": "Duncan",
"suffix": ""
}
],
"year": 1974,
"venue": "Language in Society",
"volume": "3",
"issue": "",
"pages": "161--180",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Duncan, S., On the structure of speaker-auditor interaction during speaking turns, Language in Society, vol. 3, pp. 161-180, 1974.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Independent Content Language for Integrated Text and Graphics Generation",
"authors": [
{
"first": "N",
"middle": [],
"last": "Green",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Carenini",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Kerpedjiev",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Roth",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Media",
"suffix": ""
}
],
"year": 1998,
"venue": "Proc. of Workshop on Content Visualization and Intermedia Representations at COLING and ACL '98",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Green, N., Carenini, G., Kerpedjiev, S., & Roth, S, A Media-Independent Content Language for Integrated Text and Graphics Generation, Proc. of Workshop on Content Visualization and Intermedia Representations at COLING and ACL '98, 1998.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Attention, Intentions, and the Structure of Discourse",
"authors": [
{
"first": "B",
"middle": [],
"last": "Grosz",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Sidner",
"suffix": ""
}
],
"year": 1986,
"venue": "Computational Linguistics",
"volume": "12",
"issue": "",
"pages": "175--204",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Grosz, B. & Sidner, C., Attention, Intentions, and the Structure of Discourse, Computational Linguistics, vol. 12, pp. 175-204, 1986.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Some Relationships between Body Motion and Speech, in Studies in Dyadic Communication",
"authors": [
{
"first": "A",
"middle": [],
"last": "Kendon",
"suffix": ""
}
],
"year": 1972,
"venue": "",
"volume": "",
"issue": "",
"pages": "177--210",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kendon, A., Some Relationships between Body Motion and Speech, in Studies in Dyadic Communication, A. W. Siegman and B. Pope, Eds. Elmsford, NY: Pergamon Press, 1972, pp. 177-210.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Using Plan Recognition in Human-Computer Collaboration",
"authors": [
{
"first": "N",
"middle": [],
"last": "Lesh",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Rich",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Sidner",
"suffix": ""
}
],
"year": 1999,
"venue": "Proc. of the Conference on User Modelling",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lesh, N., Rich, C., & Sidner, C., Using Plan Recognition in Human-Computer Collaboration, Proc. of the Conference on User Modelling, Banff, Canada, NY: Springer Wien, 1999.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Deictic and Emotive Communication in Animated Pedagogical Agents, in Embodied Conversational Agents",
"authors": [
{
"first": "J",
"middle": [],
"last": "Lester",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Towns",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Callaway",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Voerman",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Fitzgerald",
"suffix": ""
}
],
"year": 2000,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lester, J., Towns, S., Callaway, C., Voerman, J., & FitzGerald, P., Deictic and Emotive Communication in Animated Pedagogical Agents, in Embodied Conversational Agents, J. Cassell, J. Sullivan, et. al, Eds. Cambridge: MIT Press, 2000.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "A Collaborative Planning Model of Intentional Structure",
"authors": [
{
"first": "K",
"middle": [],
"last": "Lochbaum",
"suffix": ""
}
],
"year": 1998,
"venue": "Computational Linguistics",
"volume": "24",
"issue": "",
"pages": "525--572",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lochbaum, K., A Collaborative Planning Model of Intentional Structure, Computational Linguistics, vol. 24, pp. 525-572, 1998.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Hand and Mind: What Gestures Reveal about Thought",
"authors": [
{
"first": "D",
"middle": [],
"last": "Mcneill",
"suffix": ""
}
],
"year": 1992,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "McNeill, D., Hand and Mind: What Gestures Reveal about Thought. Chicago, IL/London, UK: The University of Chicago Press, 1992.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "COLLAGEN: A Collaboration Manager for Software Interface Agents",
"authors": [
{
"first": "C",
"middle": [],
"last": "Rich",
"suffix": ""
},
{
"first": "C",
"middle": [
"L"
],
"last": "Sidner",
"suffix": ""
}
],
"year": 1998,
"venue": "User Modeling and User-Adapted Interaction",
"volume": "8",
"issue": "",
"pages": "315--350",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rich, C. & Sidner, C. L., COLLAGEN: A Collaboration Manager for Software Interface Agents, User Modeling and User-Adapted Interaction, vol. 8, pp. 315-350, 1998.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Task-Oriented Collaboration with Embodied Agents in Virtual Worlds, in Embodied Conversational Agents",
"authors": [
{
"first": "J",
"middle": [],
"last": "Rickel",
"suffix": ""
},
{
"first": "W",
"middle": [
"L"
],
"last": "Johnson",
"suffix": ""
}
],
"year": 2000,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rickel, J. & Johnson, W. L., Task-Oriented Collaboration with Embodied Agents in Virtual Worlds, in Embodied Conversational Agents, J. Cassell, Ed. Cambridge, MA: MIT Press, 2000.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "An Artificial Discourse Language for Collaborative Negotiation",
"authors": [
{
"first": "C",
"middle": [],
"last": "Sidner",
"suffix": ""
}
],
"year": null,
"venue": "Proc. of 12th Intnl",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sidner, C., An Artificial Discourse Language for Collaborative Negotiation, Proc. of 12th Intnl.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Conf. on Artificial Intelligence (AAAI)",
"authors": [],
"year": 1994,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Conf. on Artificial Intelligence (AAAI), Seattle, WA, MIT Press, 1994.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Communicative facial displays as a new conversational modality",
"authors": [
{
"first": "A",
"middle": [],
"last": "Takeuchi",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Nagao",
"suffix": ""
}
],
"year": 1993,
"venue": "Proc. of InterCHI '93",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Takeuchi, A. & Nagao, K., Communicative facial displays as a new conversational modality, Proc. of InterCHI '93, Amsterdam, NL, ACM, 1993.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Evaluation and Integration of Speech and Pointing Gestures during Referential Understanding",
"authors": [
{
"first": "L",
"middle": [],
"last": "Thompson",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Massaro",
"suffix": ""
}
],
"year": 1986,
"venue": "Journal of Experimental Child Psychology",
"volume": "42",
"issue": "",
"pages": "144--168",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Thompson, L. and Massaro, D., Evaluation and Integration of Speech and Pointing Gestures during Referential Understanding, Journal of Experimental Child Psychology, vol. 42, pp. 144-168, 1986.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "Posture Shift Duration by DSeg and Turn",
"num": null,
"uris": null,
"type_str": "figure"
},
"FIGREF1": {
"text": "1 shows a dialogue between Rea and the user, and shows how Rea decides to generate posture shifts. This dialogue consists of two major segments: finding a house (dialogue), and showing a house (pseudo-monologue). Based on this task structure, we defined plan recipes for Collagen. The first shared discourse purpose[goal: HaveConversation] is introduced by the user before the example. Then, in utterance (1), the user introduces the main part of the conversation [goal: FindHouse]. The next goal in the agenda, [goal: IdentifyPreferredCity], should be accomplished to identify a parameter value for [goal: FindHouse]. This goal directly contributes to the current purpose, [goal: FindHouse].",
"num": null,
"uris": null,
"type_str": "figure"
},
"FIGREF2": {
"text": "In utterance (25), Rea introduces a new discourse purpose [goal : ShowHouse]. Rea, using a default rule, decides to take the initiative on this goal. At this point, Rea accesses the discourse state and confirms that a new goal is about to start. Rea judges this case as a discourse segment boundary and also a turn boundary(Case (a)). Based on this information, Rea selects a high energy posture shift. An example of Rea's high energy posture shift is shown on the right inFigure 5.2.",
"num": null,
"uris": null,
"type_str": "figure"
},
"FIGREF3": {
"text": "Rea demonstrating a low and high energy posture shift",
"num": null,
"uris": null,
"type_str": "figure"
},
"FIGREF4": {
"text": "Finding a house] < dialogue> (1) U: I'm looking for a house. (2) R: (c) Where do you want to live? (f) (3) U: I like Boston. (4) R: (c) (d) What kind of transportation access do you need? (f) (5) U: I need T access. \u2026. (23) R: (c) (d) How much storage space do you need? (f) (24) U: I need to have a storage place in the basement. (25) R: (a) (d) Let's look at 123 Elm Street. (f) (26) U: OK. [Discuss a feature of the house] (27) R: (g) Let's discuss a feature of this place. (28) R: (h) Notice the hardw ood flooring in the living room. (29) R: (h) Notice the jacuzzi. (30) R: (h) Notice the remodeled kitchen [Showing a house] <Pseudo-monologue> Example dialogue",
"num": null,
"uris": null,
"type_str": "figure"
},
"TABREF0": {
"text": "",
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null
},
"TABREF1": {
"text": "",
"type_str": "table",
"html": null,
"content": "<table><tr><td/><td>ps/s</td><td>ps/int</td><td>energy</td></tr><tr><td>inter-turn</td><td>0.140</td><td>0.268</td><td>0.742</td></tr><tr><td>intra-turn</td><td>0.022</td><td/><td>0.738</td></tr></table>",
"num": null
},
"TABREF2": {
"text": "",
"type_str": "table",
"html": null,
"content": "<table><tr><td>ps/s</td><td>ps/int</td></tr></table>",
"num": null
},
"TABREF4": {
"text": "",
"type_str": "table",
"html": null,
"content": "<table><tr><td/><td/><td/><td colspan=\"2\">Dialogue</td><td>Reaction</td></tr><tr><td/><td/><td/><td colspan=\"2\">Manager</td><td>Module (RM)</td></tr><tr><td/><td/><td/><td colspan=\"2\">(Collagen)</td></tr><tr><td/><td/><td colspan=\"2\">Understanding</td><td>Generation Module</td></tr><tr><td/><td/><td colspan=\"2\">Module</td><td>Sentence</td><td>Gesture</td></tr><tr><td/><td/><td/><td/><td>Realizer</td><td>Component</td></tr><tr><td/><td/><td>Speech</td><td colspan=\"2\">Vision</td></tr><tr><td/><td/><td>Recognition</td><td colspan=\"2\">Processing</td><td>Animation</td><td>Text to</td></tr><tr><td/><td/><td/><td/><td>Renderer</td><td>Speech</td></tr><tr><td/><td/><td>Microphone</td><td colspan=\"2\">Camera</td></tr><tr><td/><td/><td/><td/><td>Animation</td><td>Speech</td></tr><tr><td/><td/><td colspan=\"3\">Figure5.1: System architecture</td></tr><tr><td>8</td><td/><td/><td/></tr><tr><td>7</td><td/><td/><td/></tr><tr><td>6</td><td>end</td><td/><td/></tr><tr><td>5</td><td/><td/><td/></tr><tr><td>4</td><td/><td/><td/></tr><tr><td>3</td><td>start</td><td/><td/></tr><tr><td>2</td><td/><td>mid</td><td/></tr><tr><td>inter 1</td><td>DSEG</td><td>intra</td><td/></tr><tr><td/><td/><td/><td/><td>Functions</td><td>non-verbal</td></tr><tr><td/><td/><td/><td/><td>behavior cues</td></tr><tr><td/><td/><td>Discourse</td><td/><td>new segment</td><td>Posture_shift</td></tr><tr><td/><td/><td>structure</td><td/></tr><tr><td/><td/><td colspan=\"2\">Conversation</td><td>turn giving</td><td>eye_gaze &</td></tr><tr><td/><td/><td>structure</td><td/><td>(stop_gesturing</td></tr><tr><td/><td/><td/><td/><td>hand_gesture)</td></tr><tr><td/><td/><td/><td/><td>turn keeping</td><td>(look_away</td></tr><tr><td/><td/><td/><td/><td>keep_gesture)</td></tr><tr><td/><td/><td/><td/><td>turn taking</td><td>eye_gaze &</td></tr><tr><td/><td/><td/><td/><td>posture_shift</td></tr><tr><td/><td/><td>Information</td><td/><td>emphasize</td><td>eye_gaze &</td></tr><tr><td/><td/><td>structure</td><td/><td>information</td><td>beat_and</td></tr><tr><td/><td/><td/><td/><td>other_hand_gsts</td></tr></table>",
"num": null
},
"TABREF7": {
"text": "",
"type_str": "table",
"html": null,
"content": "<table><tr><td/><td/><td/><td>.3.1. For</td></tr><tr><td colspan=\"4\">example, in Case (a), the system selects a</td></tr><tr><td colspan=\"4\">posture shift with high energy, using both upper</td></tr><tr><td colspan=\"4\">and lower body. After deciding whether or not</td></tr><tr><td colspan=\"4\">Rea should shift posture and (if so) choosing a</td></tr><tr><td colspan=\"4\">kind of posture shift, Rea sends a command to</td></tr><tr><td colspan=\"4\">the Generation Module to generate a specific</td></tr><tr><td colspan=\"4\">kind of posture shift within a specific time</td></tr><tr><td colspan=\"2\">duration.</td><td/><td/></tr><tr><td>Ca se</td><td>Discourse structure information</td><td>Posture shift decision probability</td><td>Posture shift selection energy</td></tr><tr><td>g</td><td>change topic</td><td>0.84/int</td><td>high</td></tr><tr><td>h</td><td/><td/><td/></tr></table>",
"num": null
},
"TABREF8": {
"text": "",
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null
}
}
}
} |