-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
355 lines (330 loc) · 11.4 KB
/
index.html
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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8" />
<title>
Hi-ToM: A Benchmark for Evaluating Higher-Order Theory of Mind Reasoning
in Large Language Models
</title>
<link rel="shortcut icon" href="img/lit-logo.png" />
<meta
name="description"
content="Many NLP benchmarks exhibit inherent disagreements. Rather than aggregating labels, we train models directly on datasets with these disagreements. We introduce embedding-based techniques to enhance model performance on such data."
/>
<meta
name="keywords"
content="Machine Learning, dataset, classification, NLI, natural language inference, humor, sentiment analysis, emotion classification, hate speech detection, Natural Language Processing, annotation disagreement, research, EMNLP 2023 Findings, EMNLP, Deep Learning, NLP, PyTorch"
/>
<meta
name="author"
content="Naihao Deng, Siyang Liu, Xinliang Frederick Zhang, Winston Wu, Lu Wang, Rada Mihalcea"
/>
<meta name="viewport" content="width=device-width, initial-scale=1" />
<meta property="og:type" content="website" />
<meta
property="og:site_name"
content="Hi-ToM: A Benchmark for Evaluating Higher-Order Theory of Mind Reasoning in Large Language Models"
/>
<meta
property="og:image"
content="https://lit.eecs.umich.edu/annotation-embeddings-website/img/example.png"
/>
<meta property="og:image:height" content="630" />
<meta property="og:image:width" content="1200" />
<meta
property="og:title"
content="Hi-ToM: A Benchmark for Evaluating Higher-Order Theory of Mind Reasoning in Large Language Models"
/>
<meta
property="og:description"
content="Many NLP benchmarks exhibit inherent disagreements. Rather than aggregating labels, we train models directly on datasets with these disagreements. We introduce embedding-based techniques to enhance model performance on such data."
/>
<meta
property="og:url"
content="https://lit.eecs.umich.edu/annotation-embeddings-website/"
/>
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:site" content="@michigan_AI" />
<meta name="twitter:creator" content="@michigan_AI" />
<script
async
src="https://www.googletagmanager.com/gtag/js?id=G-42MFV87X10"
></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag("js", new Date());
gtag("config", "G-42MFV87X10");
</script>
<link rel="stylesheet" type="text/css" href="main.css" />
</head>
<body>
<div class="container">
<header>
<a href="https://lit.eecs.umich.edu/"
><img id="arc" src="img/lit-logo.png" alt="LIT lab logo"
/></a>
<a href="https://umich.edu/"
><img id="um" src="img/um.png" alt="University of Michigan logo"
/></a>
<h1>
Hi-ToM: A Benchmark for Evaluating Higher-Order Theory of Mind <br />
Reasoning in Large Language Models
</h1>
<ul id="quick-links">
<li><a href="https://arxiv.org/abs/2310.16755">Paper</a></li>
<li>
<a href="https://github.com/ying-hui-he/Hi-ToM_dataset">Code</a>
</li>
<li>
<a href="https://huggingface.co/umwyf/Hi-ToM_Dataset"
>Hi-ToM dataset</a
>
</li>
<li>
<a href="BibTeX.html">BibTeX Citation</a>
</li>
</ul>
</header>
<section class="section-alt">
<div class="content">
<h2>Abstract</h2>
<p id="abstract">
Theory of Mind (ToM) is the ability to reason about one's own and
others' mental states. ToM plays a critical role in the development
of intelligence, language understanding, and cognitive processes.
While previous work has primarily focused on first and second-order
ToM, we explore higher-order ToM, which involves recursive reasoning
on individuals' mental states in complex scenarios.<br /><br />
We introduce <strong>Hi-ToM</strong>, a <strong>Hi</strong>gher
<strong>O</strong>rder <strong>T</strong>heory <strong>o</strong>f
<strong>M</strong>ind benchmark. Our experimental evaluation using
various Large Language Models (LLMs) indicates a decline in
performance on higher-order ToM tasks, demonstrating the limitations
of current LLMs. We conduct a thorough analysis of different failure
cases of LLMs, and share our thoughts on the implications of our
findings on the future of NLP.
</p>
</div>
</section>
<section>
<div class="content">
<a href="https://arxiv.org/pdf/2310.16755.pdf">
<ol id="thumbnails">
<li>
<img
src="img/thumbs/1.png"
alt="thumbnail, page 1"
style="width: 75px; height: 100px"
/>
</li>
<li>
<img
src="img/thumbs/2.png"
alt="thumbnail, page 2"
style="width: 75px; height: 100px"
/>
</li>
<li>
<img
src="img/thumbs/3.png"
alt="thumbnail, page 3"
style="width: 75px; height: 100px"
/>
</li>
<li>
<img
src="img/thumbs/4.png"
alt="thumbnail, page 4"
style="width: 75px; height: 100px"
/>
</li>
<li>
<img
src="img/thumbs/5.png"
alt="thumbnail, page 5"
style="width: 75px; height: 100px"
/>
</li>
<li>
<img
src="img/thumbs/6.png"
alt="thumbnail, page 6"
style="width: 75px; height: 100px"
/>
</li>
<li>
<img
src="img/thumbs/7.png"
alt="thumbnail, page 7"
style="width: 75px; height: 100px"
/>
</li>
<li>
<img
src="img/thumbs/8.png"
alt="thumbnail, page 8"
style="width: 75px; height: 100px"
/>
</li>
<li>
<img
src="img/thumbs/9.png"
alt="thumbnail, page 9"
style="width: 75px; height: 100px"
/>
</li>
</ol>
</a>
</div>
</section>
<section>
<div class="content">
<ol id="authors">
<li>
<a href="https://ying-hui-he.github.io/">
<div class="author-img-container">
<img
src="img/authors/yinghuihe.jpg"
alt="Yinghui He profile picture"
/>
</div>
Yinghui He
</a>
</li>
<li>
<a href="https://www.linkedin.com/in/yufan-wu-a27b6b24b/">
<div class="author-img-container">
<img
src="img/authors/yufanwu.jpg"
alt="Yufan Wu profile picture"
/>
</div>
Yufan Wu
</a>
</li>
<li>
<a href="https://www.linkedin.com/in/yilin-jia-1277a1250/">
<div class="author-img-container">
<img
src="img/authors/yilin_jia.jpg"
alt="Yilin Jia profile picture"
/>
</div>
Yilin Jia
</a>
</li>
<li>
<a href="https://web.eecs.umich.edu/~mihalcea/">
<div class="author-img-container">
<img
src="img/authors/rada_mihalcea.png"
alt="Rada Mihalcea profile picture"
/>
</div>
Rada Mihalcea
</a>
</li>
<li>
<a
href="https://www.linkedin.com/in/yulong-chen-95a52614b/?originalSubdomain=uk"
>
<div class="author-img-container">
<img
src="img/authors/yulong_chen.jpg"
alt="Yulong Chen profile picture"
/>
</div>
Yulong Chen
</a>
</li>
<li>
<a href="https://dnaihao.github.io/">
<div class="author-img-container">
<img
src="img/authors/naihao_deng.jpg"
alt="Naihao Deng profile picture"
/>
</div>
Naihao Deng
</a>
</li>
</ol>
</div>
</section>
<section>
<div class="content">
<h2>Downloads</h2>
<ul id="downloads">
<li>
<a href="https://arxiv.org/pdf/2310.16755.pdf" download
><b>PDF Paper</b></a
>
</li>
<br />
<li>
<a href="https://github.com/ying-hui-he/Hi-ToM_dataset"
><b>Code</b></a
>
</li>
<br />
<a href="https://huggingface.co/datasets/Hi-ToM/Hi-ToM_Dataset"
><b>Hi-ToM dataset</b></a
>
</ul>
</div>
</section>
<section class="section-alt">
<p id="affiliation">
<a href="https://umich.edu/">
<img
id="um-vertical"
alt="University of Michigan"
src="img/um-vertical.png"
/>
</a>
</p>
</section>
<footer>
<div class="content" class="section-alt">
<h2>Acknowledgments</h2>
<p id="acknowledgments-text">
We thank the anonymous reviewers for their valuable feedback and
discussion. This paper's draft version was accepted to the
non-archival track of the ToM workshop at ICML 2023. We would also
like to extend our appreciation to the reviewers from the ToM
workshop for their feedback.
</p>
<p>
Web page inspired by the
<a href="https://lit.eecs.umich.edu/lifeqa/">LifeQA web page</a>.
</p>
</div>
</footer>
</div>
</body>
</html>
<script type="text/javascript">
function playEvidence($id, $start, $end) {
const $video = document.getElementById("example-video");
$video.pause();
document.getElementById($id).style.color = "rgb(117, 116, 116)";
function checkTime() {
if ($video.currentTime >= $end) {
$video.pause();
} else {
/* call checkTime every 1/10th second until endTime */
setTimeout(checkTime, 100);
}
}
$video.focus();
$video.currentTime = $start;
setTimeout(function () {
// to prevent `The play() request was interrupted by a call to pause().`
$video.play();
}, 150);
checkTime();
}
</script>