File size: 53,522 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 |
{
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:28:02.082725Z"
},
"title": "Generating Quantified Referring Expressions through Attention-Driven Incremental Perception",
"authors": [
{
"first": "Gordon",
"middle": [],
"last": "Briggs",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Artificial Intelligence U.S. Naval Research Laboratory Washington",
"location": {
"postCode": "20375",
"region": "DC",
"country": "USA"
}
},
"email": "gordon.briggs@nrl.navy.mil"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "We model the production of quantified referring expressions (QREs) that identity collections of visual items. A previous approach, called Perceptual Cost Pruning, modeled human QRE production using a preference-based referring expression generation algorithm, first removing facts from the input knowledge base based on a model of perceptual cost. In this paper, we present an alternative model that incrementally constructs a symbolic knowledge base through simulating human visual attention/perception from raw images. We demonstrate that this model produces the same output as Perceptual Cost Pruning. We argue that this is a more extensible approach and a step toward developing a wider range of processlevel models of human visual description.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "We model the production of quantified referring expressions (QREs) that identity collections of visual items. A previous approach, called Perceptual Cost Pruning, modeled human QRE production using a preference-based referring expression generation algorithm, first removing facts from the input knowledge base based on a model of perceptual cost. In this paper, we present an alternative model that incrementally constructs a symbolic knowledge base through simulating human visual attention/perception from raw images. We demonstrate that this model produces the same output as Perceptual Cost Pruning. We argue that this is a more extensible approach and a step toward developing a wider range of processlevel models of human visual description.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Modeling the generation of human-like referring expressions in visual contexts is an ongoing challenge in the field of natural language generation (NLG). One key aspect of this challenge is the disparity between how humans and current computational approaches operate at a process-level. Humans have perceptual and cognitive limitations; they can not mentally represent visual scenes down to the exact detail of every visual object or collection of objects. Thus, people tend to selectively attend to visual scenes in order to acquire enough information to complete their task (Yarbus, 1967) .",
"cite_spans": [
{
"start": 577,
"end": 591,
"text": "(Yarbus, 1967)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In contrast, current referring expression generation (REG) algorithms generally assume a fullyspecified, symbolic knowledge base (Van Deemter, 2016) . Essentially, these approaches abstract away the process of perception and seek to model patterns of human language primarily at the content selection phase. One task where this approach breaks down is quantified reference expression (QRE) generation. Figure 1: Examples of QRE generation task problems, based on examples from (Barr et al., 2013) . QRE tasks involve referring to collections of visual items by communicating information about the quantities contained in each collection. Initial experimental work in human-produced QREs showed regularities in responses that were not easily explained by content selection processes (Barr et al., 2013) . To illustrate this, consider the two examples of QRE problems found in Figure 1 . In Problem A, it was found that participants favored relative quantity expressions (e.g., \"the box with the most circles\") over exact number expressions (e.g., \"the box with thirty-one circles\"). In contrast, participants favored exact expressions in Problem B (Barr et al., 2013) .",
"cite_spans": [
{
"start": 129,
"end": 148,
"text": "(Van Deemter, 2016)",
"ref_id": "BIBREF14"
},
{
"start": 477,
"end": 496,
"text": "(Barr et al., 2013)",
"ref_id": "BIBREF0"
},
{
"start": 782,
"end": 801,
"text": "(Barr et al., 2013)",
"ref_id": "BIBREF0"
},
{
"start": 1147,
"end": 1166,
"text": "(Barr et al., 2013)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [
{
"start": 875,
"end": 883,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "To account for this finding, the experimenters appealed to a principle of least perceptual effort. Determining the exact number of approximately 30 objects is effortful and time-consuming, whereas determining that there are exactly three objects is quick and requires little effort. Thus, the findings can be explained by people generating exact QREs so long as determining the exact quantity did not exceed some threshold of effort.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "This explanation was validated using a computational method called Perceptual Cost Pruning, proposed previously by (Briggs and Harner, 2019 ).",
"cite_spans": [
{
"start": 115,
"end": 139,
"text": "(Briggs and Harner, 2019",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Perceptual Cost Pruning was able to model human QRE production by: (1) starting with a complete symbolic knowledge base representing the visual scene; (2) removing facts from the input knowledge base based on a model of the time cost of exact enumeration; (3) using a preference-based referring expression generation algorithm (i.e., the Incremental Algorithm) on this reduced knowledge base. We call this approach a destructive approach, wherein a full knowledge base is degraded to be sparser.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we present a constructive approach to modeling the production of QREs, wherein a sparse, symbolic scene representation is built up from nothing. Specifically, we present an alternative model that incrementally builds a symbolic knowledge base through simulating human visual attention/perception from raw images. We demonstrate that this model produces the same output as perceptual cost pruning. Furthermore, we argue that this is a more extensible approach and a step toward developing a wider range of process-level models of human visual description.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Stimuli To model the QRE task we use the stimuli described by (Barr et al., 2013) . These stimuli constitute 20 QRE generation problems, each with one target collection and two distractor collections. The items in each collection are all the same with respect to size, shape, color, and all other attributes. They are also randomly distributed in their respective containers. Thus, the total quantity of items in each collection becomes a salient feature to differentiate collections, instead of individual object attributes or collection attributes like structured arrangement (e.g., shape of the group of items). While the precise quantities for each target and distractor pair were available, the original images were not. As such, we constructed a series of 120 images, six for each problem, corresponding to the different possible target and distractor configurations. Examples of these images can be found in Figure 1 .",
"cite_spans": [
{
"start": 62,
"end": 81,
"text": "(Barr et al., 2013)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [
{
"start": 915,
"end": 923,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "The QRE Task",
"sec_num": "2"
},
{
"text": "Output In the original study (Barr et al., 2013) , participants' QREs were annotated as following into different categories, and these annotations were refined in subsequent work (Briggs and Harner, 2019) . We use the latest annotation categories, which are: Exact Number (NUM) (e.g., \"the box with 31 dots\"); Relative quantity (REL) (e.g., \"the box with the most dots\"); and Absolute description (ABS) (e.g., \"the box with dots\"). The QRE generation algorithm presented in this paper is designed to predict which of these categories is included given a particular QRE task image.",
"cite_spans": [
{
"start": 29,
"end": 48,
"text": "(Barr et al., 2013)",
"ref_id": "BIBREF0"
},
{
"start": 179,
"end": 204,
"text": "(Briggs and Harner, 2019)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The QRE Task",
"sec_num": "2"
},
{
"text": "Our model of QRE generation was developed within the ARCADIA cognitive framework (Bridewell and Bello, 2016) . ARCADIA provides an ideal framework upon which to implement a model of incremental perception and quantified description for the following reasons: (1) attention and its strategic control is the central organizing feature of the system; (2) it provides the representational flexibility necessary for modeling both an approximate number system that supports group quantity estimation and an object-tracking system that supports exact enumeration (i.e., counting); (3) it aims to implement a cognitively plausible model of human visual processing; and (4) it operates in discrete cycles, allowing for modeling the time course of perceptual and cognitive processes. In the interest of space, we omit more precise technical details regarding ARCADIA models, which can be found in other work (Bridewell and Bello, 2016; Lovett et al., 2019) .",
"cite_spans": [
{
"start": 81,
"end": 108,
"text": "(Bridewell and Bello, 2016)",
"ref_id": "BIBREF2"
},
{
"start": 898,
"end": 925,
"text": "(Bridewell and Bello, 2016;",
"ref_id": "BIBREF2"
},
{
"start": 926,
"end": 946,
"text": "Lovett et al., 2019)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "A Constructive Model of QRE Generation",
"sec_num": "3"
},
{
"text": "To produce the range of exact, relative, or absolute quantified expressions requires different forms of numerical representation at different levels of precision. Research shows that multiple forms of numerical representation underlie human number sense (Feigenson et al., 2004) . Approximate representations are obtained quickly through estimation (Barth et al., 2003) , while exact representations are obtained slowly through counting for quantities outside the subitizing range of about four items (Gelman and Gallistel, 1986) . Furthermore, evidence suggests that approximate and exact representations of quantity are obtained through different ways of deploying spatial attention (Hyde and Wood, 2011). Attention to groups results in estimation and approximate representation of quantity, while attention to individual objects underlies exact enumeration.",
"cite_spans": [
{
"start": 254,
"end": 278,
"text": "(Feigenson et al., 2004)",
"ref_id": "BIBREF7"
},
{
"start": 349,
"end": 369,
"text": "(Barth et al., 2003)",
"ref_id": "BIBREF1"
},
{
"start": 501,
"end": 529,
"text": "(Gelman and Gallistel, 1986)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Representations",
"sec_num": "3.1"
},
{
"text": "In ARCADIA, approximate quantities are represented as Gaussians with a mean, \u00b5 = n, equal to Figure 2 : A screenshot of the ARCADIA model of QRE production in mid-run. The vertically oriented red bar in the task image indicates the current scope of spatial attention as it is being swept through the image. The Working Memory display shows the current state of the symbolic scene representation. Note that the last box has not yet been attended and encoded into working memory. Also note that no exact count was obtained for box-1, unlike box-0, as the quantity estimate indicated that it was too high effort to enumerate exactly.",
"cite_spans": [],
"ref_spans": [
{
"start": 93,
"end": 101,
"text": "Figure 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Representations",
"sec_num": "3.1"
},
{
"text": "the number of items n in a group that receives attentional focus, with a standard deviation, \u03c3 = w * n, where w is the Weber fraction associated with the simulated approximate number system. Here w = 0.13, based on empirical data (Odic et al., 2013) . Exact numbers are represented by count words, and the correct sequence of count words is represented in the system as assumed knowledge.",
"cite_spans": [
{
"start": 230,
"end": 249,
"text": "(Odic et al., 2013)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Representations",
"sec_num": "3.1"
},
{
"text": "Below we give a high-level description of how our model operates. A screenshot of the model in operation can be found in Figure 2 . Videos of the model operating over example problems can also be found online. 1",
"cite_spans": [],
"ref_spans": [
{
"start": 121,
"end": 129,
"text": "Figure 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Process",
"sec_num": "3.2"
},
{
"text": "The model performs a left-to-right sweep of spatial attention. For each lightly colored square box it encounters, it encodes into the symbolic scene representation in working memory the existence of a new box of items. Likewise for each of these boxes, spatial attention is then focused on the box and a subtask that encodes information about each collection within the box is initiated. The steps of this subtask are described below.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Incremental Perception",
"sec_num": "3.2.1"
},
{
"text": "Step 1: Target/Distractor Classification: The model determines whether or not the current 1 https://osf.io/6rsg7/?view only= 034f98a2449243e28e2a593797039093 collection is the target collection by considering the background color of the box.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Incremental Perception",
"sec_num": "3.2.1"
},
{
"text": "Step 2: Quantity Estimation: The model then attends to the group of items within the box, resulting in an quantity estimate that is encoded and associated with the collection. This quantity estimate is encoded into working memory. Information is then encoded about whether or not items are present or absent (abs-desc) from the box.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Incremental Perception",
"sec_num": "3.2.1"
},
{
"text": "Step 3: Countability Judgment: Next, a number is randomly sampled from the approximate distribution encoded during the previous quantity estimation step. If this number is lower than a countability threshold value \u03c4 count , then the quantity is deemed low effort enough to exactly enumerate and the model proceeds to Step 4. Otherwise, the model determines that exact enumeration will be too high effort, and skips Step 4. Evidence from the literature on numerosity perception provides support for this notion of a countability judgment guiding enumeration strategy (Mandler and Shebo, 1982) . Furthermore, we set our countability threshold value \u03c4 count = 7 based on experimental results from (Mandler and Shebo, 1982) .",
"cite_spans": [
{
"start": 566,
"end": 591,
"text": "(Mandler and Shebo, 1982)",
"ref_id": "BIBREF11"
},
{
"start": 694,
"end": 719,
"text": "(Mandler and Shebo, 1982)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Incremental Perception",
"sec_num": "3.2.1"
},
{
"text": "Step 4: Exact Enumeration: In the case when a collection is gauged to be countable, an exact count is associated with the box (num-desc). Spatial attention is then swept downwards within the collection. Individual objects are encoded into visual short term memory, and for each new object detected, the count is incremented.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Incremental Perception",
"sec_num": "3.2.1"
},
{
"text": "Finally, after all applicable approximate and exact quantity representations are encoded for each box, the quantity information of each distractor is then compared with that of the target to ascertain information about the relative quantity of the target (rel-desc) compared with the distractors. Specifically, exact quantities are first compared if available. In the absence of exact counts, the approximate representations are compared. This is achieved by considering the Gaussian distribution that results from subtracting the Gaussian representing the distractor estimate from the Gaussian representing the target estimate. If the cumulative probability distribution of this new distribution for negative values above a threshold \u03c4 rel , this indicates that the target has a smaller quantity. Conversely, if the cumulative probability distribution for positive values is above the threshold, this indicates that the target has a larger quantity. In the scope of this paper, we set \u03c4 rel = 0.75. By comparing the target to each distractor in this manner, we determine whether the target is the collection with the most or fewest items (or neither).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Incremental Perception",
"sec_num": "3.2.1"
},
{
"text": "Having completed the process of incrementally constructing a symbolic scene representation, the model then begins the REG process. To do this we use a modified version of the the Incremental Algorithm (Dale and Reiter, 1995) . The modification is based on the one proposed by Briggs and Harner (2019) , wherein attributes with missing values are skipped. To evaluate the model, we used the best performing preference order from (Briggs and Harner, 2019) : num-desc > abs-desc > rel-desc.",
"cite_spans": [
{
"start": 201,
"end": 224,
"text": "(Dale and Reiter, 1995)",
"ref_id": "BIBREF6"
},
{
"start": 276,
"end": 300,
"text": "Briggs and Harner (2019)",
"ref_id": "BIBREF3"
},
{
"start": 428,
"end": 453,
"text": "(Briggs and Harner, 2019)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Content Selection",
"sec_num": "3.2.2"
},
{
"text": "As a proof-of-concept evaluation, we ran the model over images corresponding to the twenty unique QRE task problems modeled by Perceptual Cost Pruning by Briggs and Harner (2019) . We then compared the resulting output attribute selections to the output reported for the Perceptual Cost Pruning algorithm configured with the same attribute preference order, verifying that it was the same for all problems. The problem descriptions and corresponding model output is found in Table 1 .",
"cite_spans": [
{
"start": 154,
"end": 178,
"text": "Briggs and Harner (2019)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [
{
"start": 475,
"end": 482,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "3.3"
},
{
"text": "People can not mentally represent complex visual scenes down to the exact detail. Evidence suggests that people strategically attend to scenes to collect enough detail to complete their task. In other words, people build up sparse scene representations. Successfully modeling human performance on the QRE task requires the ability to model what pieces of information people do or do not encode. Prior work successfully modeled QRE generation through a destructive process, in which a complete scene representation is reduced according to models of perceptual cost. Here, we have presented a constructive approach to modeling QRE generation by simulating a process of incremental scene perception.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion",
"sec_num": "4"
},
{
"text": "What does the constructive, incremental approach give us over the destructive one? First, it provides a model that begins with raw images, rather than a symbolic knowledge base. Second, we argue that the constructive approach is more extensible when applied to a wider range of possible visual description or REG tasks. The destructive approach requires the development of a separate model of perceptual cost for each attribute that it considers, which for complex domains could necessitate quite complicated collections of models. In the constructive approach that we present, considerations of perceptual cost and the order in which facts are encoded are built into the perceptual process and how attention is strategically controlled. One example where a constructive approach involving incremental scene representation would be beneficial is in describing more complex visual scenes with grouped items. In contrast to the stimuli in the present work, which are randomly scattered collections of visual items, common visual scenes often contain items that are organized together in multiple groups. In some cases, visual grouping can reduce the perceptual cost of determining the total quantity of items. Visual grouping allows for people to attend to and enumerate each group separately, establishing the total number through mental arithmetic, instead of slowly counting each item one-by-one (Starkey and McCandliss, 2014; Ciccione and Dehaene, 2020) . Recent work indicates that when generating quantified descriptions of scenes with visual groups of the same cardinality, people have a tendency to omit descriptions of total quantity, and instead describe the number of groups and the number of items within each group (Briggs et al., 2020) . This finding is consistent with the idea of incremental scene representation and that knowledge about the number of groups and cardinality of each group precedes knowledge of total quantity.",
"cite_spans": [
{
"start": 1397,
"end": 1427,
"text": "(Starkey and McCandliss, 2014;",
"ref_id": "BIBREF13"
},
{
"start": 1428,
"end": 1455,
"text": "Ciccione and Dehaene, 2020)",
"ref_id": "BIBREF5"
},
{
"start": 1726,
"end": 1747,
"text": "(Briggs et al., 2020)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Approach Advantages",
"sec_num": "4.1"
},
{
"text": "We intend to use the model presented in this paper as a basis to explore how incremental perception can explain individual variation in the forms of QREs observed in (Barr et al., 2013) . While the present model waits for the scene representation to be encoded before content generation is started, it could be modified to begin the content planning phase earlier, before the scene is fully processed.",
"cite_spans": [
{
"start": 166,
"end": 185,
"text": "(Barr et al., 2013)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Future Work",
"sec_num": "4.2"
},
{
"text": "We predict that in certain QRE problems, the order in which distractors are perceived and compared with the target may affect the content of generated expressions. Specifically, consider Problem 16 described in Table 1 . The model predicts a QRE containing both a relative and absolute description of quantity. Examples of different expressions that fit this template are: (A) \"the one with some, but not the most\"; and (B) \"the one with fewer, but not the empty one\". Expressions similar to forms A and B were found when the human data were reex-amined (Briggs and Harner, 2019) . We predict that the order in which the two distractors are perceived would determine which expression form is more commonly produced. Form A would correspond to the empty distractor being perceived first, while form B would correspond to the empty distractor being perceived second.",
"cite_spans": [
{
"start": 554,
"end": 579,
"text": "(Briggs and Harner, 2019)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [
{
"start": 211,
"end": 218,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Future Work",
"sec_num": "4.2"
}
],
"back_matter": [
{
"text": "This work was supported by an NRL Karles Fellowship awarded to author and AFOSR MIPR grant F4FGA07074G001. The views expressed in this paper are solely those of the authors and should not be taken to reflect any official policy or position of the United States Government or the Department of Defense.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Generation of quantified referring expressions: evidence from experimental data",
"authors": [
{
"first": "Dale",
"middle": [],
"last": "Barr",
"suffix": ""
},
{
"first": "Raquel",
"middle": [],
"last": "Kees Van Deemter",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Fern\u00e1ndez",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the 14th European Workshop on Natural Language Generation",
"volume": "",
"issue": "",
"pages": "157--161",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dale Barr, Kees van Deemter, and Raquel Fern\u00e1ndez. 2013. Generation of quantified referring expres- sions: evidence from experimental data. In Proceed- ings of the 14th European Workshop on Natural Lan- guage Generation, pages 157-161.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "The construction of large number representations in adults",
"authors": [
{
"first": "Hilary",
"middle": [],
"last": "Barth",
"suffix": ""
},
{
"first": "Nancy",
"middle": [],
"last": "Kanwisher",
"suffix": ""
},
{
"first": "Elizabeth",
"middle": [],
"last": "Spelke",
"suffix": ""
}
],
"year": 2003,
"venue": "Cognition",
"volume": "86",
"issue": "3",
"pages": "201--221",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hilary Barth, Nancy Kanwisher, and Elizabeth Spelke. 2003. The construction of large number representa- tions in adults. Cognition, 86(3):201-221.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "A Theory of Attention for Cognitive Systems",
"authors": [
{
"first": "Will",
"middle": [],
"last": "Bridewell",
"suffix": ""
},
{
"first": "F",
"middle": [],
"last": "Paul",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Bello",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Fourth Annual Conference on Advances in Cognitive Systems",
"volume": "",
"issue": "",
"pages": "1--16",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Will Bridewell and Paul F Bello. 2016. A Theory of Attention for Cognitive Systems. In Proceedings of the Fourth Annual Conference on Advances in Cog- nitive Systems, pages 1-16, Evanston, USA.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Generating quantified referring expressions with perceptual cost pruning",
"authors": [
{
"first": "Gordon",
"middle": [],
"last": "Briggs",
"suffix": ""
},
{
"first": "Hillary",
"middle": [],
"last": "Harner",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 12th International Conference on Natural Language Generation",
"volume": "",
"issue": "",
"pages": "11--18",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gordon Briggs and Hillary Harner. 2019. Generating quantified referring expressions with perceptual cost pruning. In Proceedings of the 12th International Conference on Natural Language Generation, pages 11-18, Tokyo, Japan.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Visual grouping and pragmatic constraints in the generation of quantified descriptions",
"authors": [
{
"first": "Gordon",
"middle": [],
"last": "Briggs",
"suffix": ""
},
{
"first": "Hillary",
"middle": [],
"last": "Harner",
"suffix": ""
},
{
"first": "Sangeet",
"middle": [],
"last": "Khemlani",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 42nd Annual Virtual Meeting of the Cognitive Science Society",
"volume": "",
"issue": "",
"pages": "1008--1014",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gordon Briggs, Hillary Harner, and Sangeet Khemlani. 2020. Visual grouping and pragmatic constraints in the generation of quantified descriptions. In Pro- ceedings of the 42nd Annual Virtual Meeting of the Cognitive Science Society, pages 1008-1014.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Grouping mechanisms in numerosity perception",
"authors": [
{
"first": "Lorenzo",
"middle": [],
"last": "Ciccione",
"suffix": ""
},
{
"first": "Stanislas",
"middle": [],
"last": "Dehaene",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"DOI": [
"10.31234/osf.io/p6ryv"
]
},
"num": null,
"urls": [],
"raw_text": "Lorenzo Ciccione and Stanislas Dehaene. 2020. Grouping mechanisms in numerosity perception.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Computational interpretations of the gricean maxims in the generation of referring expressions",
"authors": [
{
"first": "Robert",
"middle": [],
"last": "Dale",
"suffix": ""
},
{
"first": "Ehud",
"middle": [],
"last": "Reiter",
"suffix": ""
}
],
"year": 1995,
"venue": "Cognitive science",
"volume": "19",
"issue": "2",
"pages": "233--263",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Robert Dale and Ehud Reiter. 1995. Computational interpretations of the gricean maxims in the gener- ation of referring expressions. Cognitive science, 19(2):233-263.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Core systems of number",
"authors": [
{
"first": "Lisa",
"middle": [],
"last": "Feigenson",
"suffix": ""
},
{
"first": "Stanislas",
"middle": [],
"last": "Dehaene",
"suffix": ""
},
{
"first": "Elizabeth",
"middle": [],
"last": "Spelke",
"suffix": ""
}
],
"year": 2004,
"venue": "Trends in Cognitive Sciences",
"volume": "8",
"issue": "",
"pages": "307--314",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lisa Feigenson, Stanislas Dehaene, and Elizabeth Spelke. 2004. Core systems of number. Trends in Cognitive Sciences, 8:307-314.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "The child's understanding of number",
"authors": [
{
"first": "Rochel",
"middle": [],
"last": "Gelman",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Charles R Gallistel",
"suffix": ""
}
],
"year": 1986,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rochel Gelman and Charles R Gallistel. 1986. The child's understanding of number. Harvard Univer- sity Press.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Spatial attention determines the nature of nonverbal number representation",
"authors": [
{
"first": "C",
"middle": [],
"last": "Daniel",
"suffix": ""
},
{
"first": "Justin",
"middle": [
"N"
],
"last": "Hyde",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Wood",
"suffix": ""
}
],
"year": 2011,
"venue": "Journal of Cognitive Neuroscience",
"volume": "23",
"issue": "",
"pages": "2336--2351",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Daniel C Hyde and Justin N Wood. 2011. Spatial at- tention determines the nature of nonverbal number representation. Journal of Cognitive Neuroscience, 23:2336-2351.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Selection enables enhancement: An integrated model of object tracking",
"authors": [
{
"first": "Andrew",
"middle": [],
"last": "Lovett",
"suffix": ""
},
{
"first": "Will",
"middle": [],
"last": "Bridewell",
"suffix": ""
},
{
"first": "Paul",
"middle": [],
"last": "Bello",
"suffix": ""
}
],
"year": 2019,
"venue": "Journal of Vision",
"volume": "19",
"issue": "14",
"pages": "1--31",
"other_ids": {
"DOI": [
"10.1167/19.14.23"
]
},
"num": null,
"urls": [],
"raw_text": "Andrew Lovett, Will Bridewell, and Paul Bello. 2019. Selection enables enhancement: An inte- grated model of object tracking. Journal of Vision, 19(14):1-31.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Subitizing: An analysis of its component processes",
"authors": [
{
"first": "George",
"middle": [],
"last": "Mandler",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Billie",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Shebo",
"suffix": ""
}
],
"year": 1982,
"venue": "Journal of Experimental Psychology: General",
"volume": "111",
"issue": "",
"pages": "1--22",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "George Mandler and Billie J Shebo. 1982. Subitizing: An analysis of its component processes. Journal of Experimental Psychology: General, 111:1-22.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Developmental change in the acuity of approximate number and area representations",
"authors": [
{
"first": "Darko",
"middle": [],
"last": "Odic",
"suffix": ""
},
{
"first": "Melissa",
"middle": [
"E"
],
"last": "Libertus",
"suffix": ""
},
{
"first": "Lisa",
"middle": [],
"last": "Feigenson",
"suffix": ""
},
{
"first": "Justin",
"middle": [],
"last": "Halberda",
"suffix": ""
}
],
"year": 2013,
"venue": "Developmental psychology",
"volume": "49",
"issue": "6",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Darko Odic, Melissa E Libertus, Lisa Feigenson, and Justin Halberda. 2013. Developmental change in the acuity of approximate number and area representa- tions. Developmental psychology, 49(6):1103.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "The emergence of \"groupitizing\" in children's numerical cognition",
"authors": [
{
"first": "S",
"middle": [],
"last": "Gillian",
"suffix": ""
},
{
"first": "Bruce",
"middle": [
"D"
],
"last": "Starkey",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Mccandliss",
"suffix": ""
}
],
"year": 2014,
"venue": "Journal of Experimental Child Psychology",
"volume": "126",
"issue": "",
"pages": "120--137",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gillian S Starkey and Bruce D McCandliss. 2014. The emergence of \"groupitizing\" in children's numerical cognition. Journal of Experimental Child Psychol- ogy, 126:120-137.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Computational models of referring: a study in cognitive science",
"authors": [
{
"first": "Kees",
"middle": [],
"last": "Van Deemter",
"suffix": ""
}
],
"year": 2016,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kees Van Deemter. 2016. Computational models of re- ferring: a study in cognitive science. MIT Press.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Eye movements during perception of complex objects",
"authors": [
{
"first": "L",
"middle": [],
"last": "Alfred",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Yarbus",
"suffix": ""
}
],
"year": 1967,
"venue": "Eye movements and vision",
"volume": "",
"issue": "",
"pages": "171--211",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Alfred L Yarbus. 1967. Eye movements during percep- tion of complex objects. In Eye movements and vi- sion, pages 171-211. Springer.",
"links": null
}
},
"ref_entries": {
"TABREF1": {
"html": null,
"text": "QRE problems and resulting model output.",
"content": "<table/>",
"num": null,
"type_str": "table"
}
}
}
} |