1 00:00:09,810 --> 00:00:12,220 hi everyone and change 2 00:00:12,840 --> 00:00:17,250 and this is mike and we're gonna talk about that big to inputs 3 00:00:17,490 --> 00:00:22,280 and first of all thank you all for attending i'll talk was 4 00:00:23,420 --> 00:00:27,120 when i don't here and this is a force time to glottic 5 00:00:27,570 --> 00:00:32,420 and i was talking to seven people about input so that nobody was kind of 6 00:00:32,420 --> 00:00:38,900 interesting stuff so i guess you have the guy is kind of interested in and 7 00:00:38,900 --> 00:00:40,280 that is really good for us 8 00:00:40,870 --> 00:00:46,000 so first of all i would like to time t-norm being because they are the 9 00:00:46,000 --> 00:00:48,020 first one that for that 10 00:00:49,200 --> 00:00:52,330 the there S all the audience as well 11 00:00:52,970 --> 00:00:57,060 who are really interested in non in using you all the languages 12 00:00:57,520 --> 00:01:02,150 and maybe last year we integrate that i would see the norm and that was 13 00:01:02,150 --> 00:01:06,990 you most were listing but we had some this solar discussion i don't it around 14 00:01:06,990 --> 00:01:10,770 ignore mailing list and things but honestly for us 15 00:01:11,370 --> 00:01:15,060 is it by testing which put how have one in the back stop 16 00:01:15,640 --> 00:01:20,080 and i would really like to thank john than the T S 17 00:01:20,680 --> 00:01:22,630 and we for the work 18 00:01:23,320 --> 00:01:27,990 and maybe let's talk that then i'll start let's talk 19 00:01:28,960 --> 00:01:31,260 you will be i'm going to talk about more about 20 00:01:31,930 --> 00:01:33,720 what are can put them at the side 21 00:01:34,170 --> 00:01:35,600 then why i 22 00:01:36,630 --> 00:01:41,500 help protect input matters that a quite and a bit of terror ticket part behind 23 00:01:41,500 --> 00:01:43,180 it and then 24 00:01:43,780 --> 00:01:47,360 the projects currently what we are working on so that you really get to know 25 00:01:47,360 --> 00:01:49,340 about more in a boat 26 00:01:49,880 --> 00:01:52,480 you predicted stuff and 27 00:01:53,020 --> 00:01:55,360 that's just for the i didn't have to the 28 00:01:56,290 --> 00:02:01,370 and if you are having any questions at any time nice feel to interrupt us 29 00:02:01,680 --> 00:02:05,540 so that we can and so at that point at so i'll be happy to 30 00:02:05,540 --> 00:02:07,040 take down the questions as well 31 00:02:07,860 --> 00:02:09,890 so all let starts 32 00:02:10,870 --> 00:02:15,030 a one of the input matters because i did this slide 33 00:02:15,570 --> 00:02:20,780 because most of you are not over know what i input like this are because 34 00:02:20,780 --> 00:02:23,030 most of the new bodies are using the 35 00:02:23,670 --> 00:02:28,710 in the this spanish keyboard or all the english keyboard or the next a keyboard 36 00:02:29,340 --> 00:02:33,310 so i thought it would be really good idea to use it to have the 37 00:02:33,310 --> 00:02:34,640 slice like this 38 00:02:35,330 --> 00:02:36,020 so 39 00:02:36,730 --> 00:02:38,530 then i put ice of input matters 40 00:02:39,200 --> 00:02:39,960 roughly 41 00:02:40,470 --> 00:02:43,840 one is kind of the rest input matters and all the rest and dispose input 42 00:02:43,840 --> 00:02:44,400 methods 43 00:02:44,820 --> 00:02:48,480 so characterbased input matters basically in D and 44 00:02:48,980 --> 00:02:55,330 cool year or vietnamese we call you at as a transliteration best input matters why 45 00:02:55,330 --> 00:02:59,980 be qualities transliteration based bit because we have the conversion between be 46 00:03:00,800 --> 00:03:06,210 ask al products or like you know products in the other are to be similar 47 00:03:06,210 --> 00:03:06,790 we can 48 00:03:07,960 --> 00:03:12,750 all the languages so that is why we called be characterbased input matters and for 49 00:03:12,750 --> 00:03:16,030 the in chinese and japanese stuff the core let's 50 00:03:16,390 --> 00:03:21,130 it's a sentence was input matters because in those input matters you do you don't 51 00:03:21,130 --> 00:03:22,680 have a 52 00:03:23,520 --> 00:03:28,700 space in between the words so it's really complex to have these such important matters 53 00:03:28,700 --> 00:03:33,440 if you see how job a japanese input methods are the japanese 54 00:03:33,760 --> 00:03:34,980 a sentence looks like 55 00:03:35,610 --> 00:03:37,250 this looks like this 56 00:03:39,820 --> 00:03:40,650 this one 57 00:03:43,360 --> 00:03:44,850 a that is a one 58 00:03:45,420 --> 00:03:47,590 this that is the whole sentence 59 00:03:47,930 --> 00:03:51,120 and is nothing but we are names in japanese 60 00:03:51,660 --> 00:03:56,770 honestly i really don't know much about japanese but mike knows here so he has 61 00:03:56,770 --> 00:04:02,700 inputted those characters if you see that on most basis in between the characters 62 00:04:03,980 --> 00:04:05,090 but there are but 63 00:04:05,700 --> 00:04:10,660 naturally they are more strict be space in between the chinese and or 64 00:04:11,470 --> 00:04:14,480 japanese stick so that becomes really hot 65 00:04:15,110 --> 00:04:18,040 to buy you japanese and chinese onto the computer 66 00:04:18,570 --> 00:04:24,240 because apparently we have only i guess thirty to a in general i'm speaking about 67 00:04:24,240 --> 00:04:30,690 but you to alphabets at such what to buy be a cactus other than the 68 00:04:30,690 --> 00:04:35,640 english or be lacking characters it's really difficult job and 69 00:04:36,070 --> 00:04:39,240 if you see right now if i use 70 00:04:40,050 --> 00:04:43,780 you know the computer in my mother tongue that is not what i think is 71 00:04:43,780 --> 00:04:48,380 moderately i of this full force at it and if you see this state of 72 00:04:48,380 --> 00:04:48,790 current 73 00:04:49,430 --> 00:04:50,660 input matters 74 00:04:51,040 --> 00:04:53,230 the state of input matters on the next all 75 00:04:53,840 --> 00:04:56,430 after typing something you see like this 76 00:04:57,760 --> 00:04:59,450 i wasn't makes its kind of face 77 00:05:00,020 --> 00:05:02,190 why was for example 78 00:05:02,640 --> 00:05:08,160 i mean you want about like norm on my own on language on the deck 79 00:05:08,160 --> 00:05:08,520 still 80 00:05:10,880 --> 00:05:15,300 i ideally it should take twenty fives you still but apparently 81 00:05:16,390 --> 00:05:21,820 it takes our own it nine you strokes and that's makes me mad why need 82 00:05:21,820 --> 00:05:28,520 to die ninety still by a word which i could buy in a english or 83 00:05:28,520 --> 00:05:29,090 be or 84 00:05:29,590 --> 00:05:32,700 i know like a keyboard profile it us 85 00:05:34,030 --> 00:05:38,570 so the predictive text is one of the way we are trying to solve that 86 00:05:38,570 --> 00:05:40,130 problem so that you that 87 00:05:40,730 --> 00:05:42,300 have to buy the less 88 00:05:42,700 --> 00:05:44,740 you get some solutions and 89 00:05:46,890 --> 00:05:49,630 maybe use this life will make this 90 00:05:49,960 --> 00:05:50,610 happy 91 00:05:51,560 --> 00:05:52,120 and 92 00:05:54,980 --> 00:05:56,460 the need for such 93 00:05:56,820 --> 00:06:00,420 that big input methods i and it dislike 94 00:06:00,940 --> 00:06:05,020 baby force today because i was a listening to keynote by 95 00:06:05,870 --> 00:06:09,820 a date and let's more when that actually arms 96 00:06:10,430 --> 00:06:16,840 i mean four buttons now but he has shown he had shown the you with 97 00:06:16,840 --> 00:06:17,160 the next 98 00:06:19,140 --> 00:06:20,020 that that's okay 99 00:06:21,350 --> 00:06:22,020 and 100 00:06:22,760 --> 00:06:29,330 he shown some more statistics about the brazil so i thought why not why not 101 00:06:29,330 --> 00:06:35,200 are working because we have like one point two one billion a population out of 102 00:06:35,200 --> 00:06:37,940 which seventy four percent are 103 00:06:38,910 --> 00:06:43,710 you can just read alright and in the language and out of reach what you 104 00:06:43,710 --> 00:06:49,800 five to six of the whole population bunch of population they can understand english i 105 00:06:49,800 --> 00:06:55,920 explicitly i did this because i've been telling on europe since last seven days and 106 00:06:55,920 --> 00:07:01,410 i met several people and the have the misconception awarding get that everyone in get 107 00:07:01,410 --> 00:07:04,160 can understand english it's really false 108 00:07:04,580 --> 00:07:09,580 in there's a out of this population five two four six percent a percent of 109 00:07:09,580 --> 00:07:14,490 the total population their billion just an english and i potentially could be one percent 110 00:07:14,490 --> 00:07:17,080 of the you open the one point two billion 111 00:07:17,630 --> 00:07:23,060 they have the you want and they use your technology they use in the operating 112 00:07:23,060 --> 00:07:28,580 system or anymore well devices and for then if you don't you do better prediction 113 00:07:28,580 --> 00:07:31,960 kind of thing they're gonna not they are not going to use E 114 00:07:33,110 --> 00:07:34,920 do you a softer for example in 115 00:07:36,410 --> 00:07:42,200 in the last year officially someone be more when you companies they sell more than 116 00:07:42,200 --> 00:07:47,010 two million and burn devices and why it's so popular in india because in and 117 00:07:47,010 --> 00:07:50,900 right you get lot of three acts as well as you get good input matters 118 00:07:52,490 --> 00:07:57,580 apparently in this room as well we use all kinds of input matters indeed more 119 00:07:57,580 --> 00:07:59,330 while or one devices 120 00:07:59,740 --> 00:08:04,030 and if you can see the dallas adjusting we have twenty two of which any 121 00:08:04,030 --> 00:08:06,360 recognise languages and i'm not just groups 122 00:08:07,240 --> 00:08:07,880 and 123 00:08:08,810 --> 00:08:11,790 if you can see that the rest of the world could be so the and 124 00:08:11,790 --> 00:08:16,870 i good languages and the users you should provide good input matters to them so 125 00:08:16,870 --> 00:08:20,960 that they can so that it will be have to present the languages 126 00:08:22,300 --> 00:08:27,890 and another point is a are we are also having the that inputs or normal 127 00:08:27,890 --> 00:08:29,590 on tablet kind of thing 128 00:08:30,110 --> 00:08:33,670 and maybe for that we need putting matter size but 129 00:08:34,450 --> 00:08:40,630 and another thing for example if you know we one language and you got really 130 00:08:40,630 --> 00:08:45,690 good in typing one language and apparently you more stuff us be more than one 131 00:08:45,690 --> 00:08:51,020 language and we know one language really but what do we really don't know the 132 00:08:51,020 --> 00:08:56,340 are the language and to typing such kind of languages it makes a really hot 133 00:08:56,890 --> 00:09:02,590 for example if you go to china and david data like really good in chinese 134 00:09:02,590 --> 00:09:06,140 what if you tell them to type in english it because makes 135 00:09:06,840 --> 00:09:09,410 because they know the language but they are not 136 00:09:09,780 --> 00:09:11,570 really good in the particular language 137 00:09:11,960 --> 00:09:14,200 so that is the need of such input matters 138 00:09:15,010 --> 00:09:15,970 and 139 00:09:16,870 --> 00:09:22,330 let's talk about how we can implement such things in fact is because to get 140 00:09:22,330 --> 00:09:23,280 this additions 141 00:09:24,000 --> 00:09:27,400 it's really hard because we have the number of words in the school you know 142 00:09:27,400 --> 00:09:27,780 was 143 00:09:28,500 --> 00:09:31,170 and how you can predict the next one 144 00:09:32,320 --> 00:09:36,260 because you really don't know that okay what i'm going to say next 145 00:09:36,590 --> 00:09:41,020 so there are two techniques what is just we use some several techniques such a 146 00:09:41,020 --> 00:09:44,600 statistical techniques and you probably did very a pretty the next one 147 00:09:45,100 --> 00:09:45,880 so 148 00:09:46,820 --> 00:09:49,150 i'll be on it as a language model 149 00:09:49,770 --> 00:09:51,650 so language model is nothing but 150 00:09:52,260 --> 00:09:53,070 of we just 151 00:09:53,500 --> 00:09:54,470 consider the problem 152 00:09:55,710 --> 00:09:59,720 in and you and language what is the probability that one what would follow before 153 00:09:59,720 --> 00:10:00,350 that word 154 00:10:00,960 --> 00:10:06,390 for example like no i'm speaking something some something about the predicted X so you 155 00:10:06,390 --> 00:10:11,540 can guess my next flawed all my neck sentences would be are something regarding the 156 00:10:11,540 --> 00:10:12,310 language model 157 00:10:13,020 --> 00:10:15,860 so similarly in probably get ready 158 00:10:16,210 --> 00:10:22,590 are incomplete us or any and but matters that does the same thing then we 159 00:10:22,590 --> 00:10:26,730 have be simple language model in that what you can see that is the number 160 00:10:26,730 --> 00:10:28,110 of a princess of words 161 00:10:29,030 --> 00:10:32,590 and divided by the number of hold what's in the language so that you get 162 00:10:32,590 --> 00:10:38,860 the probability because somewhat some sentences some words they try to getting together 163 00:10:39,370 --> 00:10:46,220 well for example i'm going so whenever i say a i then probability of the 164 00:10:46,220 --> 00:10:47,960 next what would be and 165 00:10:48,270 --> 00:10:52,320 the more score and saying it's not be exactly what 166 00:10:52,750 --> 00:10:53,880 but just you probably 167 00:10:54,720 --> 00:10:55,530 so 168 00:10:56,240 --> 00:11:01,000 if you know little about do mathematics ideally don't want to going to the that 169 00:11:01,000 --> 00:11:05,860 a good that direction what its kind of boring and will not like you much 170 00:11:06,330 --> 00:11:11,850 so the amount goes sent is in i guess in nineteen sixties or seventies had 171 00:11:11,850 --> 00:11:12,870 propose a really good 172 00:11:13,500 --> 00:11:16,010 more T V that a visa like 173 00:11:17,380 --> 00:11:19,870 if you know the idea of history 174 00:11:20,430 --> 00:11:21,040 and 175 00:11:21,830 --> 00:11:25,550 in the hysteria meant the same than you can calculate the future 176 00:11:26,310 --> 00:11:29,510 so saint at is been using machine learning technics 177 00:11:30,320 --> 00:11:30,900 but 178 00:11:32,240 --> 00:11:36,240 you can just this team next word but you can just betting the next what 179 00:11:36,940 --> 00:11:42,280 but that probability is kind of eighty percent you client base a hundred percent goes 180 00:11:42,280 --> 00:11:43,170 wide so 181 00:11:44,520 --> 00:11:50,150 because we are humans and human mind these kind of "'em" because we really don't 182 00:11:50,150 --> 00:11:50,810 know what 183 00:11:51,370 --> 00:11:52,910 we would do next 184 00:11:53,280 --> 00:11:57,170 so that makes a really hard for the text prediction 185 00:11:58,090 --> 00:12:03,990 so you probably don't do would depends on the probability of D and probably previous 186 00:12:03,990 --> 00:12:09,180 words that is the basic thing what we of what is been used in the 187 00:12:09,180 --> 00:12:15,850 text prediction so we calculate do you need honest bigrams and by bigrams unigrams is 188 00:12:15,850 --> 00:12:21,720 nothing but that's a single word by defence is nothing but set up to words 189 00:12:21,720 --> 00:12:27,240 and diagrams is nothing but a set of to us so for example know normally 190 00:12:27,240 --> 00:12:28,260 so unique it on 191 00:12:28,810 --> 00:12:29,790 well known these 192 00:12:30,470 --> 00:12:35,100 is a kind of bigram and norm is also is a trigram 193 00:12:35,920 --> 00:12:36,710 so 194 00:12:38,000 --> 00:12:42,740 you can relate such probabilities on a huge part of course say we have will 195 00:12:42,740 --> 00:12:47,870 be and so words on a given sentence so we try to calculate the unique 196 00:12:47,870 --> 00:12:49,680 don's diagrams and 197 00:12:50,260 --> 00:12:52,280 or trigrams and 198 00:12:53,030 --> 00:12:58,670 depending on that to try to calculate we try to predict the next work support 199 00:12:58,670 --> 00:12:59,330 example 200 00:13:00,640 --> 00:13:05,270 so for example containing said you have to instances 201 00:13:05,820 --> 00:13:10,290 aborting think is also norm is also and norm shall is also so there are 202 00:13:10,290 --> 00:13:11,320 two different words 203 00:13:12,840 --> 00:13:15,360 and start and stop on the team but 204 00:13:17,570 --> 00:13:20,100 space is in what you can consider the special symbol 205 00:13:20,760 --> 00:13:25,790 so that you can guess this sentence has been started and this sentence has been 206 00:13:25,790 --> 00:13:26,460 finished 207 00:13:26,940 --> 00:13:33,620 so in this example say it should know would be vocabulary in you a document 208 00:13:33,620 --> 00:13:36,120 or in your corpus here 209 00:13:37,150 --> 00:13:41,410 you will contribute start what i ease also stall and that show 210 00:13:42,810 --> 00:13:46,160 and if you want to calculate the you need a model you need ample a 211 00:13:46,160 --> 00:13:51,270 probably just for this morning is D probably you might want to consider the probability 212 00:13:51,270 --> 00:13:55,400 of you what glottic so it's one S to sixteen how com is one S 213 00:13:55,400 --> 00:14:02,070 to sixteen a because to got it is used when you understand the whole corpus 214 00:14:02,070 --> 00:14:05,950 and the number of words in the corpus to sixteen so the probabilities one it 215 00:14:05,950 --> 00:14:07,010 into sixteen 216 00:14:08,040 --> 00:14:11,940 similarly the probability of ease is do what is 217 00:14:12,960 --> 00:14:17,880 a team into sixteen this if you can apply to see mythological here so you 218 00:14:17,880 --> 00:14:18,790 can get D 219 00:14:19,650 --> 00:14:24,050 you need on model so similarly if you want to apply the same logic into 220 00:14:24,050 --> 00:14:29,030 the background model as i said trigram or the lizzie a set of keywords 221 00:14:29,480 --> 00:14:31,640 so it's so you can on 222 00:14:32,880 --> 00:14:38,380 and divided by D starts time that means that a be probably the norm using 223 00:14:38,380 --> 00:14:39,070 used 224 00:14:39,770 --> 00:14:45,700 placing this whole corpus and in the number of sentences starting with just a startup 225 00:14:45,700 --> 00:14:50,260 scene so it's politics to but see so if you apply the same logic to 226 00:14:50,260 --> 00:14:51,230 the whole sentence 227 00:14:51,980 --> 00:14:53,700 for example of 228 00:14:54,500 --> 00:14:57,440 a probability of noam got X is also meant and start 229 00:14:58,490 --> 00:15:01,710 you want to do like this you want to a lady same logic to the 230 00:15:01,710 --> 00:15:08,120 was and then you will get like probably you go text asked you into probably 231 00:15:08,120 --> 00:15:10,920 you'll ease glottic starts a single and to that end 232 00:15:11,590 --> 00:15:13,810 so it's kind of motivated by 233 00:15:14,460 --> 00:15:18,640 so that's all about the paralegal part which is kind of then that is again 234 00:15:19,210 --> 00:15:20,430 beeps and that's like 235 00:15:21,160 --> 00:15:25,170 if you don't if you get the unknown synthesis kind of thing but i to 236 00:15:25,170 --> 00:15:30,010 so how to normalise such sentences but i really don't want to will be getting 237 00:15:30,010 --> 00:15:31,110 to that complexity 238 00:15:31,730 --> 00:15:35,970 so let's talk about the projects we are working on so one of the project 239 00:15:35,970 --> 00:15:40,350 is i was type english the that do we are working on so at this 240 00:15:40,350 --> 00:15:41,730 point of time 241 00:15:42,370 --> 00:15:46,000 i didn't get to them i couldn't talk i would be 242 00:15:46,920 --> 00:15:48,200 i posted melissa 243 00:15:55,990 --> 00:15:57,960 so i tried to 244 00:15:59,600 --> 00:16:00,850 demonstrated it's 245 00:16:01,630 --> 00:16:02,070 so it 246 00:16:02,880 --> 00:16:03,350 should 247 00:16:06,150 --> 00:16:06,420 so 248 00:16:12,380 --> 00:16:17,140 i guess okay so we implemented something like that as and i was in to 249 00:16:17,140 --> 00:16:21,510 implement that it supports most language which 250 00:16:23,840 --> 00:16:28,280 can be easily transmitted weighted so it doesn't support astonish already said it doesn't support 251 00:16:28,280 --> 00:16:30,080 chinese and japanese because 252 00:16:31,020 --> 00:16:37,090 extra more complicated step to conversion to chinese characters is necessary but 253 00:16:38,380 --> 00:16:46,060 practically all other languages which can be well after consultation it's already finished are supported 254 00:16:46,750 --> 00:16:50,720 and all where directly what input is already enough 255 00:16:52,020 --> 00:16:57,610 and it users the way known input method from the M seventeen and lot of 256 00:16:57,610 --> 00:17:04,750 the so users who know D's don't a need to get used to use stuff 257 00:17:05,850 --> 00:17:12,720 and the hope is to improve typing speed a lot by getting very good predictions 258 00:17:12,720 --> 00:17:16,460 and typing on the if you look have to select the hard work 259 00:17:17,880 --> 00:17:18,560 and 260 00:17:20,830 --> 00:17:25,350 most of the prediction comes from what do you the user types it learns from 261 00:17:25,350 --> 00:17:26,170 the user input 262 00:17:26,890 --> 00:17:28,870 and it one can speed it up by 263 00:17:29,560 --> 00:17:36,980 giving some topeka text for what the user usually types to it used to time 264 00:17:36,980 --> 00:17:38,300 needed for learning 265 00:17:38,790 --> 00:17:45,490 and if i mean explain these the prediction is based on the previous two thoughts 266 00:17:45,490 --> 00:17:51,330 on that i com database and if no suitable word can lose most suitable type 267 00:17:51,330 --> 00:17:55,740 them can be found in the database it for expect to i'm spare dictionary some 268 00:17:55,740 --> 00:18:01,320 shows predictors from huntsville dictionaries and it also uses times pay for collecting minor spelling 269 00:18:01,320 --> 00:18:01,940 it was 270 00:18:04,030 --> 00:18:10,190 and currently it's implemented in the front end five what's implemented in python and this 271 00:18:10,190 --> 00:18:12,170 a database for you see collide 272 00:18:14,780 --> 00:18:15,720 and i 273 00:18:16,330 --> 00:18:19,010 why should shoulder little bit how it works 274 00:18:28,810 --> 00:18:29,370 so 275 00:18:38,150 --> 00:18:41,310 so i'm kind the german i was typing was to 276 00:18:42,110 --> 00:18:46,110 first of all i delete everything which has learned been done so far too 277 00:18:47,110 --> 00:18:48,730 to demonstrate that 278 00:18:49,760 --> 00:18:50,810 S G and it 279 00:18:52,580 --> 00:18:56,350 so if i'd type some german text 280 00:19:16,670 --> 00:19:19,250 so you see the second time i typed at 281 00:19:20,260 --> 00:19:20,580 at 282 00:19:22,100 --> 00:19:27,090 i quit just selected that typing one that and see like because it be men 283 00:19:27,090 --> 00:19:30,070 but the next about based on the previous context 284 00:19:30,740 --> 00:19:35,100 actually i this type the last about so that support on the last to be 285 00:19:35,100 --> 00:19:40,570 because i did a typing mistake and so the first say a suggestion is no 286 00:19:40,570 --> 00:19:45,340 longer if i want to delete this from the database i can selected not this 287 00:19:45,340 --> 00:19:49,830 one but this control one and sell so know this suggestion this one from the 288 00:19:49,830 --> 00:19:50,390 database 289 00:19:51,870 --> 00:19:56,250 and to speed up this learning process 290 00:19:58,630 --> 00:20:00,970 i can that we didn't some 291 00:20:01,710 --> 00:20:02,770 no not text file 292 00:20:03,930 --> 00:20:05,830 i can select lot context five 293 00:20:18,900 --> 00:20:21,000 so some example i have few have some 294 00:20:22,220 --> 00:20:25,010 some book which that the system a date 295 00:20:30,060 --> 00:20:31,280 and now if 296 00:20:37,000 --> 00:20:41,550 look at some text in this book i can easily input the 297 00:20:42,410 --> 00:20:45,550 the same text again this very little typing 298 00:21:05,030 --> 00:21:07,340 the because it are just 299 00:21:11,630 --> 00:21:15,850 you see that i'm using the german typing boost actually what i typed years english 300 00:21:15,850 --> 00:21:24,030 so for the it doesn't really matter for that items what language you are using 301 00:21:24,030 --> 00:21:30,240 you can mix the languages freely just like this with key application for the on 302 00:21:30,240 --> 00:21:31,030 the way it does 303 00:21:31,830 --> 00:21:32,630 and 304 00:21:33,450 --> 00:21:41,630 currently we still have different engines for every language but i want to much is 305 00:21:41,630 --> 00:21:45,810 in much un languages you much few engines 306 00:21:46,260 --> 00:21:49,650 to support the same them saying which is in 307 00:21:50,290 --> 00:21:51,940 on use more number of engines 308 00:21:55,830 --> 00:21:59,100 it's to something else like for a nice model to 309 00:22:04,680 --> 00:22:08,560 so you can also do the same system for practically and the 310 00:22:09,940 --> 00:22:12,340 i don't know what this means that company come out here 311 00:22:14,790 --> 00:22:15,660 and 312 00:22:19,150 --> 00:22:21,760 or queen you see that the 313 00:22:22,340 --> 00:22:25,450 suggestions the first character of suggestion 314 00:22:26,520 --> 00:22:32,020 is in i'm will actually so we see only the first john more of the 315 00:22:32,610 --> 00:22:37,660 i've typed only one jumble and the first act of that suggested lots as the 316 00:22:38,550 --> 00:22:39,780 first run most is 317 00:22:41,020 --> 00:22:41,740 korean 318 00:22:45,890 --> 00:22:49,250 okay that's the or did i think for the demonstration 319 00:22:50,680 --> 00:22:51,140 and 320 00:22:51,820 --> 00:22:52,100 cool 321 00:23:06,700 --> 00:23:07,210 well 322 00:23:09,750 --> 00:23:10,090 i think 323 00:23:16,480 --> 00:23:19,700 so the current problem solved i was 324 00:23:22,970 --> 00:23:23,240 right 325 00:23:27,390 --> 00:23:27,720 you 326 00:23:29,460 --> 00:23:33,130 you can't use the same code to go other in jeans 327 00:23:33,590 --> 00:23:38,480 or if you want to use the same girl it's really tedious so we have 328 00:23:38,480 --> 00:23:40,320 started one more project 329 00:23:40,780 --> 00:23:42,780 and if you can it's 330 00:23:43,300 --> 00:23:47,640 it's an X prediction library of which is written in the vol a so that 331 00:23:47,640 --> 00:23:51,010 you can using audit of projects as well 332 00:23:51,680 --> 00:23:56,020 just nothing but you had to well the lab is nothing but 333 00:23:56,650 --> 00:24:00,900 V handle all the key here but key variance and decline have to just subscribe 334 00:24:00,900 --> 00:24:07,030 product expectation so that once you have subscribed you'll get a prediction as it it's 335 00:24:07,900 --> 00:24:08,780 and 336 00:24:09,560 --> 00:24:13,630 the next the next service we honestly need you have 337 00:24:14,250 --> 00:24:16,890 we need help in testing 338 00:24:17,310 --> 00:24:22,340 then this additions for improvements what new features because you are you guys at the 339 00:24:22,340 --> 00:24:26,030 uses and if you have some suggestions we 340 00:24:27,360 --> 00:24:30,540 we have a happy to implement those kind of things 341 00:24:30,940 --> 00:24:37,750 and again they huntsville additional is what we are using know i honestly don't think 342 00:24:37,750 --> 00:24:42,610 nobody meant instance will dictionaries this mean or a if your C D 343 00:24:43,290 --> 00:24:46,330 i don't know i mean loss of difference billy studies 344 00:24:47,100 --> 00:24:52,520 it's kind of maybe five to six years ago somebody created them 345 00:24:53,110 --> 00:24:57,940 and all that this to something huntsville dictionaries and we would like improve grows 346 00:24:58,470 --> 00:24:59,480 and also 347 00:25:00,170 --> 00:25:02,270 a creation of we got was 348 00:25:02,600 --> 00:25:05,760 that is the thing which is really need it for us 349 00:25:06,170 --> 00:25:07,110 and 350 00:25:08,170 --> 00:25:13,200 in all what we it's really hard to get if we call was for this 351 00:25:13,200 --> 00:25:14,020 additions 352 00:25:15,010 --> 00:25:15,700 and 353 00:25:24,070 --> 00:25:29,160 so in future we might want to add some grammatical analysis as well so for 354 00:25:29,160 --> 00:25:33,960 that corpus might be interesting at the moment we are doing only this markov model 355 00:25:33,960 --> 00:25:38,370 stuff and having a big corpus doesn't actually had that much if you need to 356 00:25:38,370 --> 00:25:42,700 know you which takes like all of picky pdf for english and the prediction based 357 00:25:42,700 --> 00:25:47,430 on the simple markov model for the next about this something one out of two 358 00:25:47,430 --> 00:25:51,740 hundred fifty or one out of five hundred which isn't very good so it works 359 00:25:51,740 --> 00:25:55,100 only where at the moment if it's the 360 00:25:55,640 --> 00:25:59,750 textual on from is what the user actually uses so 361 00:26:01,940 --> 00:26:07,140 normal users don't hide and all the don't try to know complicated style like oscar 362 00:26:07,140 --> 00:26:12,280 wilde or people tend to write a better vehicle for lunch or something like this 363 00:26:12,280 --> 00:26:18,570 or the button to be could use that type just much more repetitive and having 364 00:26:18,570 --> 00:26:25,600 really learning from the user input is the markov model much more help for them 365 00:26:25,600 --> 00:26:26,700 the meeting at be corpus 366 00:26:28,270 --> 00:26:28,960 and 367 00:26:30,180 --> 00:26:33,900 and maybe that's thank you thank you only thing 368 00:26:35,600 --> 00:26:35,840 but 369 00:26:42,720 --> 00:26:48,870 you all your book on predictive implemented this are that your demonstrated also held at 370 00:26:48,870 --> 00:26:49,660 E users 371 00:26:52,070 --> 00:26:56,120 five we didn't get and if you become katie use us so far actually V 372 00:26:56,120 --> 00:27:01,070 to get pretty very little feedback so i'm is asked for test as i asked 373 00:27:01,070 --> 00:27:05,870 some of the type colleagues to tested in court some nice suggestions for improvements that 374 00:27:05,870 --> 00:27:11,900 right implemented but there wasn't that much is a feedback and that kind remember anybody 375 00:27:11,900 --> 00:27:14,820 from katie it works katie don't know so it's 376 00:27:23,240 --> 00:27:28,500 so obviously for the i think and useful and it's context but roughly make a 377 00:27:28,500 --> 00:27:30,460 production in terms of one thing keyboards 378 00:27:31,250 --> 00:27:36,650 so i'm wondering you know what you thought if you give a thought to how 379 00:27:36,650 --> 00:27:39,830 we can take this and apply it when somebody's 380 00:27:40,080 --> 00:27:45,250 using on screen keyboard results we have more general issue of how we integrate i'd 381 00:27:45,250 --> 00:27:49,090 methods with on screen keyboards but i was curious what that you had 382 00:27:49,710 --> 00:27:53,450 the county doesn't get work this on screen keyboard spot and we want to make 383 00:27:53,450 --> 00:27:57,420 it work in future this one's thinking about and that this also one of the 384 00:27:57,420 --> 00:28:01,320 reasons why on each wants to put it into a liability because the nets will 385 00:28:01,320 --> 00:28:06,220 be easier to use from an on screen keyboard and with the current implementation and 386 00:28:06,220 --> 00:28:07,040 i've just one time 387 00:28:13,040 --> 00:28:16,990 problems can see what i think it makes much more sense for actually for myself 388 00:28:16,990 --> 00:28:22,240 when i type german or english i'm typing too fast so usually for me it's 389 00:28:22,240 --> 00:28:27,750 easier to just finish typing the about instead of looking and selecting but a nice 390 00:28:27,750 --> 00:28:32,780 at that many people in india are not comfortable with the way that consultation this 391 00:28:32,780 --> 00:28:38,890 time and hard time figuring it out and so for people who use computer for 392 00:28:38,890 --> 00:28:43,560 the first time in india it's very helpful if they get some suggestions after typing 393 00:28:43,560 --> 00:28:47,560 only if you let us similar like people on the touch us clean 394 00:28:48,210 --> 00:28:49,760 have difficulties typing 395 00:28:50,380 --> 00:28:53,560 i guess that it makes me wonder a question have you thought about whether they 396 00:28:53,560 --> 00:28:59,310 should be enabled by default in some languages should just we wanted if you 397 00:29:00,090 --> 00:29:05,130 choose indian input language program should just work like this by default yes 398 00:29:05,480 --> 00:29:10,570 of the on planning like to people but one meeting the people do we need 399 00:29:10,570 --> 00:29:16,550 to fix them up to code bugs for example when you try to integrate it 400 00:29:16,550 --> 00:29:18,280 as a text of input method 401 00:29:19,160 --> 00:29:25,090 but you need to fix shootings for example you if you're typing in a say 402 00:29:25,720 --> 00:29:27,660 if you're typing something in google 403 00:29:28,090 --> 00:29:30,040 you wouldn't want to situations 404 00:29:31,040 --> 00:29:35,010 i guess i'm which means that but it it's have display some suggestions and they 405 00:29:35,010 --> 00:29:36,450 don't say it don't function 406 00:29:37,740 --> 00:29:41,370 look up table gets into the way of the good suggestion so they overlap each 407 00:29:41,370 --> 00:29:42,030 other 408 00:29:42,190 --> 00:29:42,970 so it's 409 00:29:44,840 --> 00:29:47,120 minute to switch it off and on all the time 410 00:29:49,170 --> 00:29:52,880 actually we need that would be then what that for example if you want to 411 00:29:52,880 --> 00:29:54,690 type something in those the 412 00:29:55,360 --> 00:29:59,420 and in that case as well you wouldn't require suggestions as well 413 00:30:00,990 --> 00:30:02,060 we need to do to my 414 00:30:05,850 --> 00:30:09,570 i mean indies you can actually there is maybe i to control that in some 415 00:30:09,570 --> 00:30:13,210 way now so there are these input hints that you can apply to text entry 416 00:30:13,210 --> 00:30:17,760 fields you can say i don't want it's you know this calculator i want and 417 00:30:17,760 --> 00:30:21,420 you mac stuff and this field or you can say then in your inhibit the 418 00:30:21,420 --> 00:30:25,290 on screen keyboard which you know you could then maybe imply okay and want to 419 00:30:25,290 --> 00:30:29,770 hear well prediction so maybe we can extend that technique and apply that to other 420 00:30:29,770 --> 00:30:33,130 toolkits and things that we have no good at a for something like the google 421 00:30:33,130 --> 00:30:36,690 search the field at the moment because sometimes of course if you type in the 422 00:30:36,690 --> 00:30:41,740 balls i wanted if you remain used it also for checking or whatever and how 423 00:30:41,740 --> 00:30:46,210 to find out that the user is typing into the google search for years so 424 00:30:46,210 --> 00:30:47,790 i don't know how to do that at the moment 425 00:30:53,600 --> 00:30:58,190 i think that it may do the right thing on and right i i'm not 426 00:30:58,190 --> 00:31:01,420 completely sure but i think might be maybe in is 427 00:31:01,970 --> 00:31:05,280 in H T M L so we just have to your out of expose them 428 00:31:05,280 --> 00:31:10,730 through to get the to the right place a you mean that's and they hmms 429 00:31:10,730 --> 00:31:11,790 to that page maybe 430 00:31:13,320 --> 00:31:16,410 i what we should we should listen deca see there 431 00:31:25,010 --> 00:31:27,420 okay so another questions thank you very much 432 00:31:28,080 --> 00:31:28,340 okay