WEBVTT 1 00:00:08.925 --> 00:00:23.540 Hello, my name is Anastasia Chizhikova. 2 00:00:23.540 --> 00:00:33.383 Hello, my name is Vadim Kimmelman. 3 00:00:34.651 --> 00:00:39.790 Here is the program of our presentation. 4 00:00:39.790 --> 00:00:42.426 We start by explaining 5 00:00:42.426 --> 00:00:49.099 negative headshake in sign languages. 6 00:00:49.099 --> 00:00:54.337 Then we discuss the methodology, 7 00:00:54.337 --> 00:00:58.775 then our results, 8 00:00:58.775 --> 00:01:03.480 then negative headshake in Russian Sign Language, 9 00:01:03.480 --> 00:01:06.917 including phonetic analysis. 10 00:01:06.917 --> 00:01:11.054 And then we finish with a discussion. 11 00:01:11.054 --> 00:01:16.359 How is negation in sign languages expressed? 12 00:01:16.359 --> 00:01:22.165 First by negative manual signs, like NOT. 13 00:01:22.165 --> 00:01:29.106 Second by negative non-manuals, like facial expressions and headshake. 14 00:01:29.106 --> 00:01:36.746 The most frequent marker is negative headshake. 15 00:01:36.746 --> 00:01:43.286 Almost all sign languages use negative headshake. 16 00:01:43.286 --> 00:01:55.532 But phonetic properties of headshake have barely been studied. 17 00:01:55.532 --> 00:01:58.668 Here is an example sentence 18 00:01:58.668 --> 00:02:04.341 with negative headshake in Russian Sign Language. 19 00:02:12.749 --> 00:02:17.087 What methods did we use? 20 00:02:17.087 --> 00:02:22.759 First, we searched for negative signs and sentences 21 00:02:22.759 --> 00:02:26.329 in the corpus of Russian Sign Language. 22 00:02:26.329 --> 00:02:32.169 Then we manually selected the segments 23 00:02:32.169 --> 00:02:34.504 which had negative headshake. 24 00:02:34.504 --> 00:02:41.244 Then we manually annotated boundaries of headshake 25 00:02:41.244 --> 00:02:43.213 and of manual negative signs. 26 00:02:43.213 --> 00:02:48.251 Then we cut these fragments 27 00:02:48.251 --> 00:02:58.495 and analyzed them using OpenFace. 28 00:02:58.495 --> 00:03:07.103 Here is an example of analysis with OpenFace. 29 00:03:16.313 --> 00:03:25.889 OpenFace produces head rotation coordinates. 30 00:03:25.889 --> 00:03:30.026 We analyzed the coordinates 31 00:03:30.026 --> 00:03:32.495 to find the peaks. 32 00:03:32.495 --> 00:03:40.203 We analyzed the time between peaks 33 00:03:40.203 --> 00:03:43.006 and calculated the frequency. 34 00:03:43.006 --> 00:03:54.985 We also measured the amplitude 35 00:03:54.985 --> 00:04:00.523 as the difference between the highest and lowest peaks. 36 00:04:00.523 --> 00:04:08.431 We used the frequency, amplitude, and peak count 37 00:04:08.431 --> 00:04:11.268 to analyze negative headshake. 38 00:04:11.268 --> 00:04:18.375 We found 663 examples of negative sentences. 39 00:04:18.375 --> 00:04:34.891 476 (72%) examples contained no headshake. 40 00:04:34.891 --> 00:04:42.766 But 187 examples had negative headshake 41 00:04:42.766 --> 00:04:48.271 and a manual negative sign such as 42 00:04:48.271 --> 00:04:52.609 1. The manual negative sign NEG. 43 00:04:52.609 --> 00:04:57.914 2. A sign meaning "not have" 44 00:04:57.914 --> 00:05:00.517 Russian Sign Language sign like this [NOT.HAVE]. 45 00:05:00.517 --> 00:05:05.655 3. Special negative verbs. 46 00:05:05.655 --> 00:05:09.893 For example, the sign meaning "not know" 47 00:05:09.926 --> 00:05:12.729 Russian Sign Language sign like this [NOT.KNOW]. 48 00:05:12.729 --> 00:05:18.068 We selected sentences with these manual signs 49 00:05:18.068 --> 00:05:24.941 and analyzed 68 examples phonetically. 50 00:05:24.941 --> 00:05:27.577 What did we analyze? 51 00:05:27.577 --> 00:05:30.046 We used three measurements: 52 00:05:30.046 --> 00:05:34.050 1. How many head turns? 53 00:05:34.918 --> 00:05:40.724 We found mostly 1 or 2 turns, 54 00:05:40.724 --> 00:05:44.894 but also more, up to 14 turns. 55 00:05:44.894 --> 00:05:54.637 2. We measured the amplitude 56 00:05:54.637 --> 00:06:01.578 which was 16 degrees on average. 57 00:06:01.578 --> 00:06:06.783 3. Frequency of turns 58 00:06:06.783 --> 00:06:14.290 On average, it was 7.9 turns per second. 59 00:06:15.492 --> 00:06:20.196 The three manual signs I talked about 60 00:06:20.196 --> 00:06:24.601 we analyzed phonetic properties of negative headshake 61 00:06:24.601 --> 00:06:28.571 and found no differences between the three signs. 62 00:06:28.571 --> 00:06:32.475 What are our important results? 63 00:06:32.475 --> 00:06:37.213 1. We found that negative headshake is not necessary 64 00:06:37.213 --> 00:06:39.282 to express negation. 65 00:06:39.282 --> 00:06:46.089 Negative headshake is rare in our corpus data. 66 00:06:46.089 --> 00:06:50.493 2. We found that we can 67 00:06:50.493 --> 00:06:55.098 study negation using Computer Vision. 68 00:06:55.098 --> 00:07:05.008 Output of OpenFace agrees with our observations. 69 00:07:05.008 --> 00:07:12.515 The absolute error of OpenFace 70 00:07:12.515 --> 00:07:17.887 is smaller than our measured average amplitude. 71 00:07:19.122 --> 00:07:24.160 In future, one can use our methodology 72 00:07:24.160 --> 00:07:27.630 to study other sign languages 73 00:07:27.630 --> 00:07:29.365 and compare to our results. 74 00:07:29.365 --> 00:07:38.174 We also found other headshake in our corpus data. 75 00:07:38.174 --> 00:07:44.514 In future we can study non-negative headshake 76 00:07:44.514 --> 00:07:48.852 and compare it with our results. 77 00:07:50.386 --> 00:07:51.921 That's our talk! 78 00:07:51.921 --> 00:07:56.025 If you have questions, send us an email! 79 00:07:56.025 --> 00:07:58.228 Thank you!