[Script Info] Title: [Events] Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text Dialogue: 0,0:00:05.84,0:00:07.08,Default,,0000,0000,0000,,Hi, I'm Lucie. Dialogue: 0,0:00:07.08,0:00:11.93,Default,,0000,0000,0000,,You know me from rambling about \Nnot enough language data in Wikidata, Dialogue: 0,0:00:11.93,0:00:15.80,Default,,0000,0000,0000,,and I thought instead of rambling today, \Nwhich I'll leave to Lydia later today, Dialogue: 0,0:00:16.53,0:00:20.37,Default,,0000,0000,0000,,I'll just show you a bit, or give you\Nan insight on the projects we did Dialogue: 0,0:00:20.37,0:00:25.17,Default,,0000,0000,0000,,using the data that we already have \Non Wikidata, for different causes. Dialogue: 0,0:00:25.17,0:00:28.55,Default,,0000,0000,0000,,So underserved languages \Ncompared to the keynote we just heard Dialogue: 0,0:00:28.55,0:00:32.53,Default,,0000,0000,0000,,where the person was talking about \Nunderserved as like minority languages, Dialogue: 0,0:00:32.53,0:00:35.60,Default,,0000,0000,0000,,underserved languages to me,\Nor any languages Dialogue: 0,0:00:35.60,0:00:38.76,Default,,0000,0000,0000,,that don't have \Nenough representation on the web. Dialogue: 0,0:00:39.42,0:00:40.93,Default,,0000,0000,0000,,Yeah, just to get that clear. Dialogue: 0,0:00:40.93,0:00:43.06,Default,,0000,0000,0000,,So, who am I? Dialogue: 0,0:00:43.06,0:00:45.91,Default,,0000,0000,0000,,Why am I always talking \Nabout languages on Wikidata? Dialogue: 0,0:00:45.91,0:00:47.59,Default,,0000,0000,0000,,Not sure but... Dialogue: 0,0:00:47.59,0:00:50.28,Default,,0000,0000,0000,,I'm a Computer Science PhD student Dialogue: 0,0:00:50.28,0:00:52.28,Default,,0000,0000,0000,,at the University of Southampton. Dialogue: 0,0:00:52.28,0:00:55.42,Default,,0000,0000,0000,,I'm a research intern \Nat Bloomberg in London, at the moment. Dialogue: 0,0:00:55.42,0:00:58.34,Default,,0000,0000,0000,,I'm a residence\Nat Newspeak House in London. Dialogue: 0,0:00:58.34,0:01:01.66,Default,,0000,0000,0000,,I am a researcher and project manager \Nfor the Scribe project, Dialogue: 0,0:01:01.66,0:01:03.23,Default,,0000,0000,0000,,which I'll go into in a bit, Dialogue: 0,0:01:03.23,0:01:08.53,Default,,0000,0000,0000,,and I recently got into the idea \Nof oral knowledge and oral citation. Dialogue: 0,0:01:08.53,0:01:10.17,Default,,0000,0000,0000,,Kimberly is sitting right there. Dialogue: 0,0:01:10.99,0:01:13.33,Default,,0000,0000,0000,,And then, occasionally,\NI have time to sleep Dialogue: 0,0:01:13.33,0:01:16.01,Default,,0000,0000,0000,,and do other things, but that's very rare. Dialogue: 0,0:01:16.68,0:01:18.62,Default,,0000,0000,0000,,So if you're interested\Nin any of those things, Dialogue: 0,0:01:18.62,0:01:20.02,Default,,0000,0000,0000,,come talk and speak to me. Dialogue: 0,0:01:20.02,0:01:23.48,Default,,0000,0000,0000,,Generally, this is an open presentation\Nand a few questions in between. Dialogue: 0,0:01:23.48,0:01:26.78,Default,,0000,0000,0000,,I'll run through a lot of things\Nin a very short time now. Dialogue: 0,0:01:27.41,0:01:30.11,Default,,0000,0000,0000,,Come to me afterwards\Nif you're interested in any of them. Dialogue: 0,0:01:30.64,0:01:32.17,Default,,0000,0000,0000,,Speak to me. I'm here. Dialogue: 0,0:01:32.17,0:01:35.46,Default,,0000,0000,0000,,I'm always very happy to speak to people. Dialogue: 0,0:01:35.46,0:01:39.11,Default,,0000,0000,0000,,So that's a bit of what\Nwe will talk about today. Dialogue: 0,0:01:39.11,0:01:41.48,Default,,0000,0000,0000,,So Wikidata, giving an introduction, Dialogue: 0,0:01:41.48,0:01:44.06,Default,,0000,0000,0000,,even though that's obviously\Nnot as necessary. Dialogue: 0,0:01:44.51,0:01:48.13,Default,,0000,0000,0000,,The article placeholder\Nis aimed for Wikipedia readers, Dialogue: 0,0:01:48.13,0:01:50.91,Default,,0000,0000,0000,,for Scribe which is aimed\Nat Wikipedia editors, Dialogue: 0,0:01:50.91,0:01:54.44,Default,,0000,0000,0000,,and then we have one topic of my research, Dialogue: 0,0:01:54.44,0:01:56.88,Default,,0000,0000,0000,,which is completely outside of Wikipedia Dialogue: 0,0:01:56.88,0:02:00.11,Default,,0000,0000,0000,,where we use Wikidata\Nfor question answering. Dialogue: 0,0:02:01.53,0:02:03.95,Default,,0000,0000,0000,,So just a quick rerun. Dialogue: 0,0:02:03.95,0:02:07.04,Default,,0000,0000,0000,,Why is Wikidata so cool\Nfor low-resource languages Dialogue: 0,0:02:07.04,0:02:10.82,Default,,0000,0000,0000,,where we have those unique identifiers? Dialogue: 0,0:02:10.82,0:02:13.37,Default,,0000,0000,0000,,I'm speaking to people that know that Dialogue: 0,0:02:13.37,0:02:14.93,Default,,0000,0000,0000,,much better than me even. Dialogue: 0,0:02:14.93,0:02:17.72,Default,,0000,0000,0000,,And then we have labels\Nin different languages. Dialogue: 0,0:02:17.72,0:02:21.82,Default,,0000,0000,0000,,Those can be in over,\NI think, 400 languages by now, Dialogue: 0,0:02:21.82,0:02:24.06,Default,,0000,0000,0000,,so we have a good option here Dialogue: 0,0:02:24.06,0:02:27.82,Default,,0000,0000,0000,,to reuse language\Nin different forms and capture it. Dialogue: 0,0:02:29.31,0:02:32.73,Default,,0000,0000,0000,,Yeah, so that's a little bit of me\Nrambling about Wikidata Dialogue: 0,0:02:32.73,0:02:34.88,Default,,0000,0000,0000,,because I can't stop it. Dialogue: 0,0:02:34.88,0:02:37.04,Default,,0000,0000,0000,,We compared Wikidata,\Ncompared to the native speaker, Dialogue: 0,0:02:37.04,0:02:39.11,Default,,0000,0000,0000,,so we can see, obviously, Dialogue: 0,0:02:39.11,0:02:41.57,Default,,0000,0000,0000,,there are languages\Nthat are widely spoken in the world. Dialogue: 0,0:02:41.57,0:02:43.84,Default,,0000,0000,0000,,There's Chinese, Hindi, or Arabic, Dialogue: 0,0:02:43.84,0:02:46.64,Default,,0000,0000,0000,,but then very low coverage on Wikidata. Dialogue: 0,0:02:48.00,0:02:50.13,Default,,0000,0000,0000,,Then the opposite. Dialogue: 0,0:02:50.13,0:02:52.59,Default,,0000,0000,0000,,Sorry, I have the Dutch\Nand the Swedish community Dialogue: 0,0:02:52.59,0:02:54.88,Default,,0000,0000,0000,,which was super active in Wikidata, Dialogue: 0,0:02:54.88,0:02:58.06,Default,,0000,0000,0000,,which is really cool, \Nand that just points out Dialogue: 0,0:02:58.06,0:03:01.33,Default,,0000,0000,0000,,that even though we have \Na low number of speakers, Dialogue: 0,0:03:01.33,0:03:06.81,Default,,0000,0000,0000,,we can have a big impact if people \Nare very active in the communities, Dialogue: 0,0:03:06.81,0:03:09.00,Default,,0000,0000,0000,,which is really nice and really good. Dialogue: 0,0:03:09.00,0:03:13.60,Default,,0000,0000,0000,,But also let's try to equal\Nthat graph out in the future. Dialogue: 0,0:03:14.56,0:03:18.57,Default,,0000,0000,0000,,So, cool. So now we have \Nall this language data in Wikidata. Dialogue: 0,0:03:18.57,0:03:22.28,Default,,0000,0000,0000,,We have low-resource Wikipedias, \Nso we thought, what can we do? Dialogue: 0,0:03:22.28,0:03:27.46,Default,,0000,0000,0000,,Well, my undergrad supervisor\Nis sitting here, Dialogue: 0,0:03:27.46,0:03:31.07,Default,,0000,0000,0000,,and we worked back then\Nin the golden days, Dialogue: 0,0:03:31.07,0:03:33.62,Default,,0000,0000,0000,,on something called\Nthe article placeholder Dialogue: 0,0:03:34.73,0:03:39.37,Default,,0000,0000,0000,,which takes triples from Wikidata\Nand displays it on Wikipedia. Dialogue: 0,0:03:39.37,0:03:41.57,Default,,0000,0000,0000,,And that's pretty much \Nrelatively straight forward. Dialogue: 0,0:03:41.57,0:03:46.30,Default,,0000,0000,0000,,So you just take the content of Wikidata,\Ndisplay it on Wikipedia Dialogue: 0,0:03:46.30,0:03:49.33,Default,,0000,0000,0000,,to attract more readers \Nand then eventually more editors Dialogue: 0,0:03:49.33,0:03:51.33,Default,,0000,0000,0000,,in the different low-resource languages. Dialogue: 0,0:03:51.33,0:03:53.13,Default,,0000,0000,0000,,They are dynamically generated, Dialogue: 0,0:03:53.13,0:03:55.55,Default,,0000,0000,0000,,so they're not like stubs or bot articles Dialogue: 0,0:03:55.55,0:04:00.17,Default,,0000,0000,0000,,that then flood the Wikipedia \Nso people can edit them. Dialogue: 0,0:04:00.17,0:04:02.42,Default,,0000,0000,0000,,It's basically a starting point. Dialogue: 0,0:04:02.42,0:04:04.55,Default,,0000,0000,0000,,And we thought, \Nwell, we have that content, Dialogue: 0,0:04:04.55,0:04:08.57,Default,,0000,0000,0000,,and we have that knowledge\Nsomewhere already, which is Wikidata. Dialogue: 0,0:04:08.57,0:04:11.60,Default,,0000,0000,0000,,It's often already in the languages,\Nbut they don't have articles, Dialogue: 0,0:04:11.60,0:04:15.14,Default,,0000,0000,0000,,so at least give them\Nthe insight into the information. Dialogue: 0,0:04:15.14,0:04:19.22,Default,,0000,0000,0000,,The article placeholders are live \Non 14 low-resource Wikipedias. Dialogue: 0,0:04:20.04,0:04:21.77,Default,,0000,0000,0000,,If you are a Wikipedia community, Dialogue: 0,0:04:21.77,0:04:24.80,Default,,0000,0000,0000,,if you are part of a Wikipedia community\Nand interested in it, Dialogue: 0,0:04:24.80,0:04:26.11,Default,,0000,0000,0000,,let us know. Dialogue: 0,0:04:27.88,0:04:30.08,Default,,0000,0000,0000,,And then I went into research, Dialogue: 0,0:04:30.08,0:04:32.77,Default,,0000,0000,0000,,and I got stuck with\Nthe article placeholder, though, Dialogue: 0,0:04:32.77,0:04:36.04,Default,,0000,0000,0000,,so we started to look into \Ntext generation from Wikidata Dialogue: 0,0:04:36.04,0:04:38.06,Default,,0000,0000,0000,,for Wikipedia and low-resource languages. Dialogue: 0,0:04:38.06,0:04:39.96,Default,,0000,0000,0000,,And text generation is really interesting Dialogue: 0,0:04:39.96,0:04:43.31,Default,,0000,0000,0000,,because in research it was at that point \Nwhen we started the project Dialogue: 0,0:04:43.31,0:04:46.02,Default,,0000,0000,0000,,completely only focused on English, Dialogue: 0,0:04:46.02,0:04:48.88,Default,,0000,0000,0000,,which is a bit pointless in my experience Dialogue: 0,0:04:48.88,0:04:51.44,Default,,0000,0000,0000,,because, I mean, you have a lot of people\Nwho write in English, Dialogue: 0,0:04:51.44,0:04:55.35,Default,,0000,0000,0000,,but then what we need is people \Nwho write in those low-source languages. Dialogue: 0,0:04:55.35,0:04:59.42,Default,,0000,0000,0000,,And our starting point was that,\Nlooking at triples on Wikipedia Dialogue: 0,0:04:59.42,0:05:01.60,Default,,0000,0000,0000,,is not exactly the nicest thing. Dialogue: 0,0:05:01.60,0:05:03.68,Default,,0000,0000,0000,,I mean, as much as I love\Nthe article placeholder, Dialogue: 0,0:05:03.68,0:05:06.33,Default,,0000,0000,0000,,it's not exactly \Nwhat you want to see you or expect Dialogue: 0,0:05:06.33,0:05:07.96,Default,,0000,0000,0000,,when you open a Wikipedia page. Dialogue: 0,0:05:07.96,0:05:09.59,Default,,0000,0000,0000,,So we try to generate text. Dialogue: 0,0:05:09.59,0:05:11.77,Default,,0000,0000,0000,,We use this beautiful\Nneural network model, Dialogue: 0,0:05:11.77,0:05:13.44,Default,,0000,0000,0000,,where we encode Wikidata triples. Dialogue: 0,0:05:13.44,0:05:15.76,Default,,0000,0000,0000,,If you're interested more\Nin the technical parts, Dialogue: 0,0:05:15.76,0:05:16.97,Default,,0000,0000,0000,,come and talk to me. Dialogue: 0,0:05:16.97,0:05:21.82,Default,,0000,0000,0000,,And so, realistically,\Nwith neural text generation, Dialogue: 0,0:05:21.82,0:05:23.75,Default,,0000,0000,0000,,you can generate one or two sentences Dialogue: 0,0:05:23.75,0:05:27.53,Default,,0000,0000,0000,,before it completely scrambles\Nand becomes useless. Dialogue: 0,0:05:27.53,0:05:32.66,Default,,0000,0000,0000,,So we've generated one sentence\Nthat describes the topic of the triple. Dialogue: 0,0:05:32.66,0:05:35.60,Default,,0000,0000,0000,,And so this, for example, is Arabic. Dialogue: 0,0:05:35.60,0:05:38.62,Default,,0000,0000,0000,,We generate the sentence about Marrakesh, Dialogue: 0,0:05:38.62,0:05:40.66,Default,,0000,0000,0000,,where it just describes the city. Dialogue: 0,0:05:42.17,0:05:45.68,Default,,0000,0000,0000,,So for that, then, we tested this-- Dialogue: 0,0:05:45.68,0:05:49.33,Default,,0000,0000,0000,,So we did studies, obviously,\Nto test if our approach works, Dialogue: 0,0:05:49.33,0:05:52.48,Default,,0000,0000,0000,,and if it makes sense, to use such things. Dialogue: 0,0:05:52.48,0:05:55.66,Default,,0000,0000,0000,,And because we are \Nvery application-focused, Dialogue: 0,0:05:55.66,0:05:58.73,Default,,0000,0000,0000,,we tested it with actual \NWikipedia readers and editors. Dialogue: 0,0:05:58.73,0:06:01.30,Default,,0000,0000,0000,,So, first, we tested it\Nwith Wikipedia readers Dialogue: 0,0:06:01.30,0:06:03.02,Default,,0000,0000,0000,,in Arabic and Esperanto-- Dialogue: 0,0:06:03.02,0:06:06.17,Default,,0000,0000,0000,,so use cases with Arabic and Esperanto. Dialogue: 0,0:06:07.64,0:06:12.71,Default,,0000,0000,0000,,And we can see that our model\Ncan generate sentences Dialogue: 0,0:06:12.71,0:06:14.49,Default,,0000,0000,0000,,that are very fluent Dialogue: 0,0:06:14.49,0:06:18.05,Default,,0000,0000,0000,,and that feel very much--\Nsurprisingly, a lot, actually-- Dialogue: 0,0:06:18.05,0:06:19.64,Default,,0000,0000,0000,,like Wikipedia sentences. Dialogue: 0,0:06:19.64,0:06:22.71,Default,,0000,0000,0000,,So it picks up, so we train on,\Nfor example, for Arabic, Dialogue: 0,0:06:22.71,0:06:26.47,Default,,0000,0000,0000,,we train on Arabic with the idea to say Dialogue: 0,0:06:26.47,0:06:29.88,Default,,0000,0000,0000,,we want to keep\Nthe cultural context of that language Dialogue: 0,0:06:29.88,0:06:32.98,Default,,0000,0000,0000,,and not let it influence Dialogue: 0,0:06:32.98,0:06:35.30,Default,,0000,0000,0000,,from other languages\Nthat have higher coverage. Dialogue: 0,0:06:36.15,0:06:38.40,Default,,0000,0000,0000,,And then we did a study\Nwith Wikipedia editors Dialogue: 0,0:06:38.40,0:06:41.08,Default,,0000,0000,0000,,because in the end the article placeholder\Nis just a starting point Dialogue: 0,0:06:41.08,0:06:42.52,Default,,0000,0000,0000,,for people to start editing, Dialogue: 0,0:06:42.52,0:06:43.57,Default,,0000,0000,0000,,and we try to measure Dialogue: 0,0:06:43.57,0:06:45.95,Default,,0000,0000,0000,,how much of the sentences\Nwould they reuse. Dialogue: 0,0:06:45.95,0:06:48.75,Default,,0000,0000,0000,,How much is useful for them, basically, Dialogue: 0,0:06:48.75,0:06:51.20,Default,,0000,0000,0000,,and you can see \Nthat there is a high number of reuse, Dialogue: 0,0:06:51.20,0:06:54.88,Default,,0000,0000,0000,,especially in Esperanto\Nwhen we test with editors. Dialogue: 0,0:06:55.86,0:07:01.15,Default,,0000,0000,0000,,And finally, we did also\Nqualitative interviews Dialogue: 0,0:07:01.15,0:07:05.03,Default,,0000,0000,0000,,with Wikipedia editors \Nacross six languages. Dialogue: 0,0:07:05.03,0:07:07.68,Default,,0000,0000,0000,,I think we had \Nabout ten people we interviewed. Dialogue: 0,0:07:08.68,0:07:12.26,Default,,0000,0000,0000,,And we tried to get\Nmore of an understanding Dialogue: 0,0:07:12.26,0:07:15.31,Default,,0000,0000,0000,,what's a human perspective\Non those generated sentences. Dialogue: 0,0:07:15.31,0:07:18.06,Default,,0000,0000,0000,,So now we can have \Na very quantified way of saying, Dialogue: 0,0:07:18.06,0:07:19.28,Default,,0000,0000,0000,,yeah, they are good, Dialogue: 0,0:07:19.28,0:07:21.28,Default,,0000,0000,0000,,but we wanted to see Dialogue: 0,0:07:21.28,0:07:22.78,Default,,0000,0000,0000,,how's the interaction Dialogue: 0,0:07:22.78,0:07:25.51,Default,,0000,0000,0000,,and especially with whatever\Nalways happens Dialogue: 0,0:07:25.51,0:07:30.34,Default,,0000,0000,0000,,in neural machine translation\Nand neural text generations, Dialogue: 0,0:07:30.34,0:07:33.97,Default,,0000,0000,0000,,that you have those missing word tokens\Nwhich we put as "rare" in there. Dialogue: 0,0:07:33.97,0:07:38.86,Default,,0000,0000,0000,,So that's the example sentences we used.\NAll of them are in Marrakesh. Dialogue: 0,0:07:38.86,0:07:42.15,Default,,0000,0000,0000,,So we wanted to see how much \Nare people bothered by it, Dialogue: 0,0:07:42.15,0:07:43.20,Default,,0000,0000,0000,,what's the quality, Dialogue: 0,0:07:43.20,0:07:45.35,Default,,0000,0000,0000,,what are the things\Nthat point out to them, Dialogue: 0,0:07:45.35,0:07:50.08,Default,,0000,0000,0000,,and we can see that the mistakes\Nby the networks like those red tokens Dialogue: 0,0:07:50.08,0:07:51.42,Default,,0000,0000,0000,,are often just ignored. Dialogue: 0,0:07:53.08,0:07:56.08,Default,,0000,0000,0000,,There is this interesting factor\Nthat because we didn't tell them Dialogue: 0,0:07:56.08,0:08:00.64,Default,,0000,0000,0000,,where this happens,\Nwhere we got the sentences from-- Dialogue: 0,0:08:00.64,0:08:03.68,Default,,0000,0000,0000,,because it was on a user page of mine Dialogue: 0,0:08:03.68,0:08:05.88,Default,,0000,0000,0000,,but it looked like it was on a Wikipedia, Dialogue: 0,0:08:05.88,0:08:07.42,Default,,0000,0000,0000,,people just trusted. Dialogue: 0,0:08:07.42,0:08:09.00,Default,,0000,0000,0000,,And I think that's very important Dialogue: 0,0:08:09.00,0:08:13.35,Default,,0000,0000,0000,,when we look into those kinds\Nof research directions that we look into, Dialogue: 0,0:08:13.35,0:08:16.13,Default,,0000,0000,0000,,we cannot override \Nthis trust into Wikipedia. Dialogue: 0,0:08:16.13,0:08:20.46,Default,,0000,0000,0000,,So if we work with Wikipedians\Nand Wikipedia itself, Dialogue: 0,0:08:20.46,0:08:23.24,Default,,0000,0000,0000,,if we take things from,\Nfor example, Wikidata, Dialogue: 0,0:08:23.24,0:08:26.40,Default,,0000,0000,0000,,that's good\Nbecause it's also human-curated. Dialogue: 0,0:08:26.40,0:08:31.05,Default,,0000,0000,0000,,But when we start\Nwith artificial intelligence projects, Dialogue: 0,0:08:31.05,0:08:34.68,Default,,0000,0000,0000,,where you have to be really careful\Nwhat we actually expose people to Dialogue: 0,0:08:34.68,0:08:37.95,Default,,0000,0000,0000,,because they just trust\Nthe information that we give them. Dialogue: 0,0:08:38.91,0:08:42.57,Default,,0000,0000,0000,,So we could see, for example,\Nin the Arabic version, Dialogue: 0,0:08:42.57,0:08:45.48,Default,,0000,0000,0000,,it gave the wrong location for Marrakesh, Dialogue: 0,0:08:45.48,0:08:47.77,Default,,0000,0000,0000,,and people, even the people I interviewed Dialogue: 0,0:08:47.77,0:08:50.33,Default,,0000,0000,0000,,that we're living in Marrakesh \Ndidn't pick up on that, Dialogue: 0,0:08:50.33,0:08:54.09,Default,,0000,0000,0000,,because it's on Wikipedia, \Nso it should be fine, right? Dialogue: 0,0:08:54.09,0:08:55.12,Default,,0000,0000,0000,,(chuckles) Dialogue: 0,0:08:55.12,0:08:56.34,Default,,0000,0000,0000,,Yeah. Dialogue: 0,0:08:57.68,0:09:00.75,Default,,0000,0000,0000,,We found there was a magical threshold\Nfor the lengths of the generated text, Dialogue: 0,0:09:00.75,0:09:02.00,Default,,0000,0000,0000,,so that's something we found, Dialogue: 0,0:09:02.00,0:09:05.25,Default,,0000,0000,0000,,especially in comparison\Nwith the content translation tool, Dialogue: 0,0:09:05.25,0:09:08.08,Default,,0000,0000,0000,,where you have a long \Nautomatically generated text, Dialogue: 0,0:09:08.08,0:09:12.13,Default,,0000,0000,0000,,and people were complaining\Nthat content translation was very hard Dialogue: 0,0:09:12.13,0:09:15.61,Default,,0000,0000,0000,,because you're just doing post-editing,\Nyou don't have the creativity. Dialogue: 0,0:09:15.61,0:09:19.33,Default,,0000,0000,0000,,There are other remarks \Non content translation I usually make-- Dialogue: 0,0:09:19.33,0:09:20.71,Default,,0000,0000,0000,,I'll skip them for now. Dialogue: 0,0:09:22.40,0:09:25.23,Default,,0000,0000,0000,,So that one sentence was helpful Dialogue: 0,0:09:25.23,0:09:30.36,Default,,0000,0000,0000,,because even if we've made mistakes,\Npeople were still willing to fix them Dialogue: 0,0:09:30.36,0:09:34.13,Default,,0000,0000,0000,,because it's a very short \Nintervenience [in that]. Dialogue: 0,0:09:34.13,0:09:37.95,Default,,0000,0000,0000,,And then, finally, \Na lot of people pointed out, Dialogue: 0,0:09:37.95,0:09:40.20,Default,,0000,0000,0000,,that it was particularly good \Nfor a new editor, Dialogue: 0,0:09:40.20,0:09:42.08,Default,,0000,0000,0000,,so for them to have a starting point, Dialogue: 0,0:09:42.08,0:09:44.08,Default,,0000,0000,0000,,to have those triples, to have a sentence, Dialogue: 0,0:09:44.08,0:09:46.15,Default,,0000,0000,0000,,so they have something to start from. Dialogue: 0,0:09:46.15,0:09:48.72,Default,,0000,0000,0000,,So after all those interviews were done, Dialogue: 0,0:09:48.72,0:09:51.99,Default,,0000,0000,0000,,as I go, that's very interesting. Dialogue: 0,0:09:51.99,0:09:54.26,Default,,0000,0000,0000,,What else can we do with that knowledge? Dialogue: 0,0:09:54.26,0:09:58.95,Default,,0000,0000,0000,,And so we started a new project, \Nexactly because there weren't enough yet. Dialogue: 0,0:09:58.95,0:10:02.31,Default,,0000,0000,0000,,And the new project we have\Nis called Scribe, Dialogue: 0,0:10:02.31,0:10:07.46,Default,,0000,0000,0000,,and Scribe focuses on new editors\Nthat want to write a new article, Dialogue: 0,0:10:07.46,0:10:09.66,Default,,0000,0000,0000,,and particularly people\Nwho haven't written Dialogue: 0,0:10:09.66,0:10:11.26,Default,,0000,0000,0000,,an article on Wikipedia yet, Dialogue: 0,0:10:11.26,0:10:14.28,Default,,0000,0000,0000,,and specifically also\Non low-resource languages. Dialogue: 0,0:10:15.13,0:10:18.91,Default,,0000,0000,0000,,So the idea is that-- \Nthat's the pixel version of me. Dialogue: 0,0:10:19.80,0:10:21.17,Default,,0000,0000,0000,,All my slides are basically Dialogue: 0,0:10:21.17,0:10:24.24,Default,,0000,0000,0000,,references to people in this room,\Nwhich I really love. Dialogue: 0,0:10:24.24,0:10:25.75,Default,,0000,0000,0000,,It feels like I'm home again. Dialogue: 0,0:10:27.00,0:10:30.88,Default,,0000,0000,0000,,So, yeah, I want to write a new article, Dialogue: 0,0:10:30.88,0:10:33.57,Default,,0000,0000,0000,,but I don't know where to start\Nas a new editor, Dialogue: 0,0:10:33.57,0:10:36.71,Default,,0000,0000,0000,,and so we have this project Scribe. Dialogue: 0,0:10:36.71,0:10:41.37,Default,,0000,0000,0000,,Scribe is a profession\Nor was the name of someone Dialogue: 0,0:10:41.37,0:10:45.26,Default,,0000,0000,0000,,with the profession of writing\Nin ancient Egypt. Dialogue: 0,0:10:47.08,0:10:52.87,Default,,0000,0000,0000,,So the Scribe project's idea\Nis that we want to give people, basically, Dialogue: 0,0:10:52.87,0:10:55.71,Default,,0000,0000,0000,,a hand when they start\Nwriting their first articles. Dialogue: 0,0:10:55.71,0:10:57.75,Default,,0000,0000,0000,,So give them a skeleton, Dialogue: 0,0:10:57.75,0:11:01.04,Default,,0000,0000,0000,,give them a skeleton that's based\Non their language Wikipedia, Dialogue: 0,0:11:01.04,0:11:05.33,Default,,0000,0000,0000,,instead of just translating the content\Nfrom another language Wikipedia. Dialogue: 0,0:11:05.33,0:11:10.39,Default,,0000,0000,0000,,So the first thing we want to do\Nis plan section titles, Dialogue: 0,0:11:10.39,0:11:13.64,Default,,0000,0000,0000,,then select references for each section, Dialogue: 0,0:11:13.64,0:11:15.97,Default,,0000,0000,0000,,ideally in the local Wikipedia language, Dialogue: 0,0:11:15.97,0:11:19.95,Default,,0000,0000,0000,,and then summarize those references\Nto give a starting point to write. Dialogue: 0,0:11:21.40,0:11:25.31,Default,,0000,0000,0000,,For the project, we have \Na Wikimedia Foundation project grant. Dialogue: 0,0:11:25.31,0:11:27.57,Default,,0000,0000,0000,,So it just started. Dialogue: 0,0:11:27.57,0:11:30.67,Default,,0000,0000,0000,,Some of you are very open\Nto feedback, in general. Dialogue: 0,0:11:30.67,0:11:35.17,Default,,0000,0000,0000,,That was the very first\Nnot so beautiful layout, Dialogue: 0,0:11:35.17,0:11:36.83,Default,,0000,0000,0000,,but just for you to get an overview. Dialogue: 0,0:11:36.83,0:11:39.64,Default,,0000,0000,0000,,So there is this idea \Nof collecting references, Dialogue: 0,0:11:39.64,0:11:42.95,Default,,0000,0000,0000,,images from comments, section titles. Dialogue: 0,0:11:42.95,0:11:45.62,Default,,0000,0000,0000,,And so the main things \Nwe want to use Wikidata for Dialogue: 0,0:11:45.62,0:11:47.85,Default,,0000,0000,0000,,is the sections. Dialogue: 0,0:11:47.86,0:11:51.40,Default,,0000,0000,0000,,So, basically, we want to see\Nwhat are articles Dialogue: 0,0:11:51.40,0:11:55.22,Default,,0000,0000,0000,,on similar topics \Nalready existing in your language, Dialogue: 0,0:11:55.22,0:11:58.35,Default,,0000,0000,0000,,so we can understand\Nhow the language community Dialogue: 0,0:11:58.35,0:12:02.22,Default,,0000,0000,0000,,decided on structuring articles. Dialogue: 0,0:12:02.22,0:12:06.17,Default,,0000,0000,0000,,And then we look\Nfor the images, obviously, Dialogue: 0,0:12:06.17,0:12:10.48,Default,,0000,0000,0000,,where Wikidata also\Nis a good point to go through. Dialogue: 0,0:12:12.55,0:12:16.24,Default,,0000,0000,0000,,And then we made\Na prettier interface for it Dialogue: 0,0:12:16.24,0:12:18.42,Default,,0000,0000,0000,,because we decided to go mobile first. Dialogue: 0,0:12:18.42,0:12:21.28,Default,,0000,0000,0000,,So most of communities\Nthat we aim to work with Dialogue: 0,0:12:21.28,0:12:24.51,Default,,0000,0000,0000,,are very heavy on mobile editing. Dialogue: 0,0:12:24.51,0:12:29.80,Default,,0000,0000,0000,,And so we do this mobile-first focus. Dialogue: 0,0:12:30.23,0:12:34.06,Default,,0000,0000,0000,,And then, it also forces us\Nto break down into steps Dialogue: 0,0:12:34.06,0:12:37.00,Default,,0000,0000,0000,,which eventually will lead to, \Nyeah, I don't know, Dialogue: 0,0:12:37.00,0:12:39.44,Default,,0000,0000,0000,,a step-by-step guide \Non how to write a new article. Dialogue: 0,0:12:39.44,0:12:43.06,Default,,0000,0000,0000,,So an editor comes,\Nthey can select section headers Dialogue: 0,0:12:43.06,0:12:46.63,Default,,0000,0000,0000,,based on existing articles\Nin their language, Dialogue: 0,0:12:46.63,0:12:49.15,Default,,0000,0000,0000,,write one section at a time, Dialogue: 0,0:12:49.15,0:12:54.13,Default,,0000,0000,0000,,switch between the sections, \Nand select references for each section. Dialogue: 0,0:12:55.60,0:12:59.05,Default,,0000,0000,0000,,Yeah, so the idea is that\Nwe will have an easier editing experience, Dialogue: 0,0:12:59.05,0:13:00.68,Default,,0000,0000,0000,,especially for new editors, Dialogue: 0,0:13:00.68,0:13:05.08,Default,,0000,0000,0000,,to keep them in-- \Nintegrate Wikidata information Dialogue: 0,0:13:05.08,0:13:08.28,Default,,0000,0000,0000,,and [inaudible] images \Nfrom Wikimedia Commons as well. Dialogue: 0,0:13:09.73,0:13:12.11,Default,,0000,0000,0000,,If you're interested in Scribe, Dialogue: 0,0:13:12.11,0:13:15.13,Default,,0000,0000,0000,,I'm working together\Non this project with Hady. Dialogue: 0,0:13:15.13,0:13:19.31,Default,,0000,0000,0000,,There is a lot of things online, Dialogue: 0,0:13:19.31,0:13:23.24,Default,,0000,0000,0000,,but then also just come and talk to us. Dialogue: 0,0:13:23.24,0:13:25.86,Default,,0000,0000,0000,,Also, if you're editing \Na low-resource Wikipedia, Dialogue: 0,0:13:25.86,0:13:28.61,Default,,0000,0000,0000,,we're still looking\Nfor people to interview Dialogue: 0,0:13:28.61,0:13:31.57,Default,,0000,0000,0000,,because we're trying to emulate-- Dialogue: 0,0:13:31.57,0:13:33.88,Default,,0000,0000,0000,,we're trying to emulate as much as we can Dialogue: 0,0:13:33.88,0:13:36.75,Default,,0000,0000,0000,,what people already experience,\Nor they already edit. Dialogue: 0,0:13:36.75,0:13:38.63,Default,,0000,0000,0000,,I'm not big on Wikipedia editing. Dialogue: 0,0:13:38.63,0:13:40.51,Default,,0000,0000,0000,,Also, my native language is German. Dialogue: 0,0:13:40.51,0:13:43.58,Default,,0000,0000,0000,,So I need a lot of input from editors Dialogue: 0,0:13:43.58,0:13:48.08,Default,,0000,0000,0000,,that want to tell me\Nwhat they need, what they want, Dialogue: 0,0:13:48.08,0:13:51.14,Default,,0000,0000,0000,,where they think this project can go. Dialogue: 0,0:13:51.14,0:13:54.59,Default,,0000,0000,0000,,And if you are into Wikidata,\Nalso come and talk to me, please. Dialogue: 0,0:13:55.73,0:13:57.89,Default,,0000,0000,0000,,Okay, so that's all the projects Dialogue: 0,0:13:57.89,0:14:01.88,Default,,0000,0000,0000,,or most of the projects we did\Ninside the Wikimedia world. Dialogue: 0,0:14:01.88,0:14:05.78,Default,,0000,0000,0000,,And I want to give you one \Nshort overview of what's happening Dialogue: 0,0:14:05.78,0:14:10.42,Default,,0000,0000,0000,,on my end of research,\Naround Wikidata as well. Dialogue: 0,0:14:14.29,0:14:15.95,Default,,0000,0000,0000,,So I was part of a project Dialogue: 0,0:14:15.95,0:14:17.82,Default,,0000,0000,0000,,that works a lot with question answering, Dialogue: 0,0:14:17.82,0:14:20.46,Default,,0000,0000,0000,,and I don't know too much\Nabout question answering, Dialogue: 0,0:14:20.46,0:14:23.88,Default,,0000,0000,0000,,but what I do know a lot about\Nis knowledge graphs and multilinguality. Dialogue: 0,0:14:23.88,0:14:25.68,Default,,0000,0000,0000,,So, basically, what we wanted to do Dialogue: 0,0:14:25.68,0:14:29.96,Default,,0000,0000,0000,,is we have a question answering system\Nthat gets a question from a user, Dialogue: 0,0:14:29.96,0:14:35.77,Default,,0000,0000,0000,,and we wanted to select a knowledge graph\Nthat can answer the question best. Dialogue: 0,0:14:35.77,0:14:40.15,Default,,0000,0000,0000,,And again, we focused on \Nmultilingual question answering system. Dialogue: 0,0:14:40.15,0:14:45.75,Default,,0000,0000,0000,,So if I want to ask something about Bach,\Nfor example, in Spanish and French-- Dialogue: 0,0:14:45.75,0:14:48.42,Default,,0000,0000,0000,,because that's the two languages\NI know best-- Dialogue: 0,0:14:48.42,0:14:52.03,Default,,0000,0000,0000,,then what knowledge graph has the data Dialogue: 0,0:14:52.03,0:14:53.95,Default,,0000,0000,0000,,to actually answer those questions. Dialogue: 0,0:14:55.16,0:14:59.26,Default,,0000,0000,0000,,So what we did was we found a method\Nto rank knowledge graphs, Dialogue: 0,0:15:00.60,0:15:04.80,Default,,0000,0000,0000,,based on the metadata of language, Dialogue: 0,0:15:04.80,0:15:08.17,Default,,0000,0000,0000,,that appears on the knowledge graph, Dialogue: 0,0:15:08.17,0:15:09.51,Default,,0000,0000,0000,,[which is split] by class. Dialogue: 0,0:15:09.51,0:15:11.46,Default,,0000,0000,0000,,And then we look for each class Dialogue: 0,0:15:11.46,0:15:14.44,Default,,0000,0000,0000,,into what languages are covered best, Dialogue: 0,0:15:14.44,0:15:18.17,Default,,0000,0000,0000,,and then depending on the question,\Ncan suggest a knowledge graph. Dialogue: 0,0:15:19.00,0:15:22.51,Default,,0000,0000,0000,,From the big knowledge graphs\Nwe looked into Dialogue: 0,0:15:22.51,0:15:25.08,Default,,0000,0000,0000,,and that are very known and widely used, Dialogue: 0,0:15:25.08,0:15:28.24,Default,,0000,0000,0000,,Wikidata covers the most languages\Nover all knowledge graphs, Dialogue: 0,0:15:28.24,0:15:31.75,Default,,0000,0000,0000,,and we used a test bed. Dialogue: 0,0:15:31.75,0:15:35.57,Default,,0000,0000,0000,,So we'd use a benchmark dataset\Ncalled [CALD], Dialogue: 0,0:15:35.57,0:15:39.35,Default,,0000,0000,0000,,which we then translated-- \Nwhich was originally for DBpedia. Dialogue: 0,0:15:39.35,0:15:41.88,Default,,0000,0000,0000,,We translated it\Nfor those five knowledge graphs Dialogue: 0,0:15:41.88,0:15:43.55,Default,,0000,0000,0000,,into [SPARQL] questions. Dialogue: 0,0:15:43.55,0:15:49.82,Default,,0000,0000,0000,,And then we gave that to a crowd \Nand looked into which knowledge graph Dialogue: 0,0:15:49.82,0:15:54.68,Default,,0000,0000,0000,,has the best answers\Nfor each of those [SPARQL] queries. Dialogue: 0,0:15:54.68,0:15:59.37,Default,,0000,0000,0000,,And overall, the crowd workers\Npreferred Wikidata's answers Dialogue: 0,0:15:59.37,0:16:01.64,Default,,0000,0000,0000,,because they are very precise, Dialogue: 0,0:16:02.89,0:16:05.02,Default,,0000,0000,0000,,they are in most of the languages Dialogue: 0,0:16:05.02,0:16:06.53,Default,,0000,0000,0000,,that the others don't cover, Dialogue: 0,0:16:07.62,0:16:10.97,Default,,0000,0000,0000,,and they are not \Nas repetitive or redundant Dialogue: 0,0:16:10.97,0:16:12.48,Default,,0000,0000,0000,,as the [inaudible]. Dialogue: 0,0:16:12.48,0:16:16.68,Default,,0000,0000,0000,,So just to make a quick recap \Non the whole topic Dialogue: 0,0:16:16.68,0:16:19.82,Default,,0000,0000,0000,,of Wikidata and the future and languages. Dialogue: 0,0:16:19.82,0:16:23.62,Default,,0000,0000,0000,,So we can say that Wikidata\Nis already widely used Dialogue: 0,0:16:23.62,0:16:27.91,Default,,0000,0000,0000,,for numerous applications in Wikipedia, Dialogue: 0,0:16:27.91,0:16:30.08,Default,,0000,0000,0000,,and then outside Wikipedia for research. Dialogue: 0,0:16:30.08,0:16:33.97,Default,,0000,0000,0000,,So what I talked about\Nis just the things I do research on, Dialogue: 0,0:16:33.97,0:16:35.97,Default,,0000,0000,0000,,but there is still so much more. Dialogue: 0,0:16:35.97,0:16:38.95,Default,,0000,0000,0000,,So there is machine translation\Nusing knowledge graphs, Dialogue: 0,0:16:38.95,0:16:40.95,Default,,0000,0000,0000,,there is rule mining\Nover knowledge graphs, Dialogue: 0,0:16:40.95,0:16:43.53,Default,,0000,0000,0000,,its entity linking in text. Dialogue: 0,0:16:43.53,0:16:47.17,Default,,0000,0000,0000,,There is so much more research\Nhappening at the moment, Dialogue: 0,0:16:47.17,0:16:50.88,Default,,0000,0000,0000,,and Wikidata is more and more\Ngetting popular for usage of it. Dialogue: 0,0:16:50.88,0:16:54.64,Default,,0000,0000,0000,,So I think we are at a very good stage Dialogue: 0,0:16:54.64,0:16:57.59,Default,,0000,0000,0000,,to push and connect the communities. Dialogue: 0,0:16:58.64,0:17:02.97,Default,,0000,0000,0000,,Yeah, to get the best \Nfrom both sides, basically. Dialogue: 0,0:17:03.51,0:17:04.77,Default,,0000,0000,0000,,Thank you very much. Dialogue: 0,0:17:04.77,0:17:07.86,Default,,0000,0000,0000,,If you want to have a look\Nat any of those projects, Dialogue: 0,0:17:07.86,0:17:09.28,Default,,0000,0000,0000,,they are there, Dialogue: 0,0:17:09.28,0:17:10.71,Default,,0000,0000,0000,,my slides are in Commons already. Dialogue: 0,0:17:10.71,0:17:14.80,Default,,0000,0000,0000,,If you want to read any of the papers, \NI think all of them are open access. Dialogue: 0,0:17:14.80,0:17:16.26,Default,,0000,0000,0000,,If you can't find any of them, Dialogue: 0,0:17:16.26,0:17:18.77,Default,,0000,0000,0000,,write me an email \Nand I send it to you immediately. Dialogue: 0,0:17:18.77,0:17:20.70,Default,,0000,0000,0000,,Thank you very much. Dialogue: 0,0:17:20.70,0:17:22.40,Default,,0000,0000,0000,,(applause) Dialogue: 0,0:17:25.74,0:17:28.13,Default,,0000,0000,0000,,(moderator) Okay, \Nare there any questions? Dialogue: 0,0:17:28.13,0:17:31.77,Default,,0000,0000,0000,,- (moderator) I'll come around.\N- (person 1) Shall I come to you? Dialogue: 0,0:17:34.79,0:17:36.37,Default,,0000,0000,0000,,(person 1) Hi Lucie, thank you so much, Dialogue: 0,0:17:36.37,0:17:38.46,Default,,0000,0000,0000,,I'm so glad to see\Nyou taking this forward. Dialogue: 0,0:17:38.46,0:17:40.68,Default,,0000,0000,0000,,Now I'm really curious about Scribe. Dialogue: 0,0:17:41.51,0:17:43.51,Default,,0000,0000,0000,,The example here within our university Dialogue: 0,0:17:43.51,0:17:46.06,Default,,0000,0000,0000,,was that the idea that the person says, Dialogue: 0,0:17:46.06,0:17:47.54,Default,,0000,0000,0000,,"This is a university." Dialogue: 0,0:17:47.54,0:17:49.02,Default,,0000,0000,0000,,And then you go to the key data Dialogue: 0,0:17:49.02,0:17:51.93,Default,,0000,0000,0000,,and say, "Oh gosh! \NUniversities have places Dialogue: 0,0:17:51.93,0:17:54.11,Default,,0000,0000,0000,,and presidents, and I don't know what," Dialogue: 0,0:17:54.11,0:17:57.77,Default,,0000,0000,0000,,that you're using these as the parts, \Nfor telling the person what to do. Dialogue: 0,0:17:57.77,0:18:00.84,Default,,0000,0000,0000,,So, basically, the idea\Nis that someone says, Dialogue: 0,0:18:00.84,0:18:02.82,Default,,0000,0000,0000,,"I want to write about Nile University." Dialogue: 0,0:18:02.82,0:18:07.04,Default,,0000,0000,0000,,We look into Nile University's\NWikidata item, Dialogue: 0,0:18:07.04,0:18:09.82,Default,,0000,0000,0000,,and let's say-- I work a lot with Arabic-- Dialogue: 0,0:18:09.82,0:18:13.24,Default,,0000,0000,0000,,so let's say we then go\Nin Arabic Wikipedia, Dialogue: 0,0:18:13.24,0:18:17.04,Default,,0000,0000,0000,,so we can make a grid, basically, Dialogue: 0,0:18:17.04,0:18:19.37,Default,,0000,0000,0000,,of all items that are around\NNile University. Dialogue: 0,0:18:19.37,0:18:23.06,Default,,0000,0000,0000,,So there are also universities,\Nthere are also universities in Cairo, Dialogue: 0,0:18:23.06,0:18:25.48,Default,,0000,0000,0000,,or there are also universities\Nin Egypt, stuff like that, Dialogue: 0,0:18:25.48,0:18:27.35,Default,,0000,0000,0000,,or they have similar topics. Dialogue: 0,0:18:27.35,0:18:32.53,Default,,0000,0000,0000,,So we can look into\Nall the similar items on Wikidata, Dialogue: 0,0:18:32.53,0:18:36.33,Default,,0000,0000,0000,,and if they already have \Na Wikipedia entry in Arabic Wikipedia, Dialogue: 0,0:18:36.33,0:18:38.61,Default,,0000,0000,0000,,we can look at the section titles. Dialogue: 0,0:18:38.61,0:18:41.31,Default,,0000,0000,0000,,- (person 1) (gasps)\N- Exactly, and then we can make basically, Dialogue: 0,0:18:41.31,0:18:46.37,Default,,0000,0000,0000,,the most common way \Nabout writing about a university Dialogue: 0,0:18:46.37,0:18:50.00,Default,,0000,0000,0000,,in Cairo on Arabic Wikipedia. Dialogue: 0,0:18:50.00,0:18:52.70,Default,,0000,0000,0000,,- Yeah, so that's the--\N- (person 1) Thank you, [inaudible]. Dialogue: 0,0:18:56.88,0:18:59.55,Default,,0000,0000,0000,,(person 2) Hi, thank you so much\Nfor your inspiring talk. Dialogue: 0,0:18:59.55,0:19:04.80,Default,,0000,0000,0000,,I was wondering if this would work\Nfor languages in Incubator? Dialogue: 0,0:19:04.80,0:19:10.62,Default,,0000,0000,0000,,Like, I work with really low,\Nlow, low, low-resource languages Dialogue: 0,0:19:10.62,0:19:16.46,Default,,0000,0000,0000,,and this thing about doing it mobile\Nwould be a huge thing, Dialogue: 0,0:19:16.46,0:19:20.02,Default,,0000,0000,0000,,because in many communities\Nthey only have phones, not laptops. Dialogue: 0,0:19:20.02,0:19:22.02,Default,,0000,0000,0000,,So, would it work? Dialogue: 0,0:19:22.02,0:19:26.08,Default,,0000,0000,0000,,So I think, to an extent-- Dialogue: 0,0:19:26.08,0:19:32.05,Default,,0000,0000,0000,,so the general structure, the skeleton\Nof the application would work. Dialogue: 0,0:19:32.05,0:19:35.28,Default,,0000,0000,0000,,Two things that we're thinking about\Na lot at the moment Dialogue: 0,0:19:35.28,0:19:37.08,Default,,0000,0000,0000,,for exactly those use cases is, Dialogue: 0,0:19:37.08,0:19:39.97,Default,,0000,0000,0000,,how much would we want,\Nfor example, to say, Dialogue: 0,0:19:39.97,0:19:44.53,Default,,0000,0000,0000,,if there are no articles \Non a similar topic in your Wikipedia, Dialogue: 0,0:19:44.53,0:19:46.93,Default,,0000,0000,0000,,how much do we want it\Nto get it from other Wikipedias. Dialogue: 0,0:19:46.93,0:19:49.75,Default,,0000,0000,0000,,And that's why I'm basically \Ndoing those interviews at the moment, Dialogue: 0,0:19:49.75,0:19:51.42,Default,,0000,0000,0000,,because I try to understand Dialogue: 0,0:19:51.42,0:19:54.57,Default,,0000,0000,0000,,how much people already look\Nat other language Wikipedias Dialogue: 0,0:19:54.57,0:19:57.04,Default,,0000,0000,0000,,to make the structure of an article. Dialogue: 0,0:19:57.04,0:19:58.80,Default,,0000,0000,0000,,Are they generally equal Dialogue: 0,0:19:58.80,0:20:01.63,Default,,0000,0000,0000,,or do they differ a lot\Nbased on cultural context? Dialogue: 0,0:20:01.63,0:20:04.31,Default,,0000,0000,0000,,So that would be something to consider, Dialogue: 0,0:20:04.31,0:20:06.64,Default,,0000,0000,0000,,but there is a possibility to say, Dialogue: 0,0:20:06.64,0:20:09.55,Default,,0000,0000,0000,,we take everything \Nfrom all the language Wikipedias Dialogue: 0,0:20:09.55,0:20:12.04,Default,,0000,0000,0000,,and then make an average, basically. Dialogue: 0,0:20:12.04,0:20:14.97,Default,,0000,0000,0000,,And the other problem is referencing. Dialogue: 0,0:20:14.97,0:20:16.46,Default,,0000,0000,0000,,So that's something we find. Dialogue: 0,0:20:16.46,0:20:20.73,Default,,0000,0000,0000,,We make it very convenient\Nbecause we use a lot of Arabic, Dialogue: 0,0:20:20.73,0:20:24.17,Default,,0000,0000,0000,,and Arabic actually has the problem\Nthat there are a lot of references, Dialogue: 0,0:20:24.17,0:20:28.79,Default,,0000,0000,0000,,but they are very little used\Nor not widely used in Wikipedia. Dialogue: 0,0:20:29.26,0:20:31.57,Default,,0000,0000,0000,,That's not true, obviously,\Nfor all languages, Dialogue: 0,0:20:31.57,0:20:34.10,Default,,0000,0000,0000,,and that's something\NI'd be very interested-- Dialogue: 0,0:20:34.10,0:20:35.18,Default,,0000,0000,0000,,like, let's talk. Dialogue: 0,0:20:35.18,0:20:36.68,Default,,0000,0000,0000,,That's what I'm trying to say, Dialogue: 0,0:20:36.68,0:20:39.17,Default,,0000,0000,0000,,I'd be very interested \Non your perspective on it Dialogue: 0,0:20:39.17,0:20:41.68,Default,,0000,0000,0000,,because I'd like to know, yeah Dialogue: 0,0:20:41.68,0:20:43.75,Default,,0000,0000,0000,,what do you think about referencing Dialogue: 0,0:20:43.75,0:20:45.46,Default,,0000,0000,0000,,done from English or any other language. Dialogue: 0,0:20:45.46,0:20:46.84,Default,,0000,0000,0000,,(person 2) Have you ever tried-- Dialogue: 0,0:20:46.84,0:20:51.88,Default,,0000,0000,0000,,what we do is we normally\Nreference to interviews we have. Dialogue: 0,0:20:51.88,0:20:55.60,Default,,0000,0000,0000,,We put them in our repository,\Ninstitutional repository, Dialogue: 0,0:20:55.60,0:20:59.57,Default,,0000,0000,0000,,because these languages \Ndon't have written references, Dialogue: 0,0:20:59.57,0:21:03.24,Default,,0000,0000,0000,,and I feel like \Nthat is the way to go, but-- Dialogue: 0,0:21:03.24,0:21:06.91,Default,,0000,0000,0000,,I'm currently also--\NKimberly and I are discussing a lot. Dialogue: 0,0:21:06.91,0:21:10.93,Default,,0000,0000,0000,,We made a session on Wikimania\Non oral knowledge and oral citations. Dialogue: 0,0:21:10.93,0:21:14.14,Default,,0000,0000,0000,,Yeah, we should hang out \Nand have a long conversation. Dialogue: 0,0:21:14.14,0:21:15.62,Default,,0000,0000,0000,,(laughs) Dialogue: 0,0:21:18.31,0:21:22.04,Default,,0000,0000,0000,,(person 3) So [Michael Davignon], \Nwe'll talk about medium size, Dialogue: 0,0:21:22.04,0:21:23.91,Default,,0000,0000,0000,,which is probably around ten people, Dialogue: 0,0:21:23.91,0:21:27.75,Default,,0000,0000,0000,,so it's medium for Briton Wikipedia. Dialogue: 0,0:21:27.75,0:21:30.60,Default,,0000,0000,0000,,And I'm wondering if we can use Scribe, Dialogue: 0,0:21:31.53,0:21:34.88,Default,,0000,0000,0000,,how to find a common plan\Nthe other way around Dialogue: 0,0:21:34.88,0:21:37.77,Default,,0000,0000,0000,,for existing article\Nto find [the outer layers], Dialogue: 0,0:21:37.77,0:21:39.57,Default,,0000,0000,0000,,that's supposed to be the best plan, Dialogue: 0,0:21:39.57,0:21:42.13,Default,,0000,0000,0000,,but I'm not aware of more or less Dialogue: 0,0:21:42.13,0:21:44.71,Default,,0000,0000,0000,,[inaudible] \Nimprovement existing article. Dialogue: 0,0:21:46.79,0:21:49.44,Default,,0000,0000,0000,,I think there's-- Dialogue: 0,0:21:49.44,0:21:50.80,Default,,0000,0000,0000,,I forgot the name, I think, Dialogue: 0,0:21:50.80,0:21:53.79,Default,,0000,0000,0000,,[Diego] in the Wikimedia Foundation\Nresearch team, Dialogue: 0,0:21:53.79,0:21:58.41,Default,,0000,0000,0000,,who's working a lot at the moment\Nwith section headings. Dialogue: 0,0:21:58.41,0:22:01.42,Default,,0000,0000,0000,,But, yes, generally, the idea is the same. Dialogue: 0,0:22:01.42,0:22:04.64,Default,,0000,0000,0000,,So instead of using them\Nto make an average Dialogue: 0,0:22:04.64,0:22:07.26,Default,,0000,0000,0000,,you could say, \Nthis is not like the average, Dialogue: 0,0:22:08.17,0:22:09.68,Default,,0000,0000,0000,,That's very possible, yeah. Dialogue: 0,0:22:14.75,0:22:18.33,Default,,0000,0000,0000,,(person 4) Hi, Lucy. I'm Erica Azzellini\Nfrom Wiki Movement, Brazil, Dialogue: 0,0:22:18.33,0:22:20.13,Default,,0000,0000,0000,,and I'm very-- Dialogue: 0,0:22:20.13,0:22:21.86,Default,,0000,0000,0000,,(Érica) Oh, can you hear me? Dialogue: 0,0:22:21.86,0:22:24.68,Default,,0000,0000,0000,,So, I'm Érica Azzellini\Nfrom Wiki Movement Brazil, Dialogue: 0,0:22:24.68,0:22:26.56,Default,,0000,0000,0000,,and I'm really impressed with your work Dialogue: 0,0:22:26.57,0:22:29.15,Default,,0000,0000,0000,,because it's really in sync Dialogue: 0,0:22:29.15,0:22:32.54,Default,,0000,0000,0000,,with what we've been working on in Brazil\Nwith the Mbabel tool. Dialogue: 0,0:22:32.54,0:22:33.95,Default,,0000,0000,0000,,I don't know if you heard about it? Dialogue: 0,0:22:33.95,0:22:36.00,Default,,0000,0000,0000,,- Not yet. \N- (Érica) It's a tool that we use Dialogue: 0,0:22:36.02,0:22:38.44,Default,,0000,0000,0000,,to automatically\Ngenerate Wikipedia entries Dialogue: 0,0:22:38.44,0:22:42.24,Default,,0000,0000,0000,,using Wikidata information \Nin a simple way Dialogue: 0,0:22:42.24,0:22:46.51,Default,,0000,0000,0000,,that can be replicated \Non other Wikipedia languages. Dialogue: 0,0:22:46.51,0:22:48.95,Default,,0000,0000,0000,,So we've been working\Non Portuguese mainly, Dialogue: 0,0:22:48.95,0:22:51.86,Default,,0000,0000,0000,,and we're trying to get\Non English Wikipedia tools, Dialogue: 0,0:22:51.86,0:22:56.20,Default,,0000,0000,0000,,but it can be replicated\Non any language, basically, Dialogue: 0,0:22:56.20,0:22:58.46,Default,,0000,0000,0000,,and I think then we could talk about it. Dialogue: 0,0:22:58.46,0:23:00.46,Default,,0000,0000,0000,,Absolutely, it will be super interesting Dialogue: 0,0:23:00.46,0:23:03.26,Default,,0000,0000,0000,,because the article placeholder\Nis an extension already, Dialogue: 0,0:23:03.26,0:23:06.13,Default,,0000,0000,0000,,so it might be worth \Nto integrate your efforts Dialogue: 0,0:23:06.13,0:23:07.95,Default,,0000,0000,0000,,into the existing extension. Dialogue: 0,0:23:07.95,0:23:12.62,Default,,0000,0000,0000,,Lydia is also fully for it,\Nand... (laughs) Dialogue: 0,0:23:12.62,0:23:13.93,Default,,0000,0000,0000,,And then because-- Dialogue: 0,0:23:13.93,0:23:17.04,Default,,0000,0000,0000,,so one of the problems--\N[Marius] correct me if I'm wrong-- Dialogue: 0,0:23:17.04,0:23:20.31,Default,,0000,0000,0000,,we had was that\Narticle placeholder doesn't scale Dialogue: 0,0:23:20.31,0:23:22.24,Default,,0000,0000,0000,,as well as it should. Dialogue: 0,0:23:22.24,0:23:24.86,Default,,0000,0000,0000,,So article placeholder\Nis not in Portuguese Dialogue: 0,0:23:24.86,0:23:28.54,Default,,0000,0000,0000,,because we're always afraid\Nit will break everything, correct? Dialogue: 0,0:23:29.46,0:23:32.29,Default,,0000,0000,0000,,And then [Marius] is just taking a pause. Dialogue: 0,0:23:32.29,0:23:35.42,Default,,0000,0000,0000,,- (Érica) Yeah, you should be careful.\N- Don't want to say anything about this. Dialogue: 0,0:23:35.42,0:23:38.95,Default,,0000,0000,0000,,But, yeah, we should connect\Nbecause I'd be super interested to see Dialogue: 0,0:23:38.95,0:23:42.04,Default,,0000,0000,0000,,how you solve those issues\Nand how it works for you. Dialogue: 0,0:23:42.04,0:23:45.31,Default,,0000,0000,0000,,(Érica) I'm going to present \Non the second section Dialogue: 0,0:23:45.31,0:23:48.35,Default,,0000,0000,0000,,of the learning talk about this project \Nthat we've been developing, Dialogue: 0,0:23:48.35,0:23:50.62,Default,,0000,0000,0000,,and we've been using it \Non [Glenwyck] initiatives Dialogue: 0,0:23:50.62,0:23:52.44,Default,,0000,0000,0000,,and education projects already. Dialogue: 0,0:23:52.44,0:23:54.48,Default,,0000,0000,0000,,- Perfect.\N- (Érica) So let's do that. Dialogue: 0,0:23:54.48,0:23:56.44,Default,,0000,0000,0000,,Yeah, absolutely let's chat. Dialogue: 0,0:23:57.22,0:23:58.27,Default,,0000,0000,0000,,(moderator) Cool. Dialogue: 0,0:23:58.27,0:24:00.37,Default,,0000,0000,0000,,Some other questions on your projects? Dialogue: 0,0:24:02.46,0:24:06.82,Default,,0000,0000,0000,,(person 5) Hi, my name is [Alan], \Nand I think this is extremely cool. Dialogue: 0,0:24:06.82,0:24:09.17,Default,,0000,0000,0000,,I had a few questions about Dialogue: 0,0:24:09.17,0:24:13.11,Default,,0000,0000,0000,,generating Wiki sentences\Nfrom neural networks. Dialogue: 0,0:24:13.11,0:24:16.02,Default,,0000,0000,0000,,- Yeah.\N- (person 5) So I've come across Dialogue: 0,0:24:16.02,0:24:19.24,Default,,0000,0000,0000,,another project\Nthat was attempting to do this, Dialogue: 0,0:24:19.24,0:24:23.02,Default,,0000,0000,0000,,and it was essentially using \N[triples input and sentences output], Dialogue: 0,0:24:23.02,0:24:25.51,Default,,0000,0000,0000,,and it was able \Nto generate very fluent sentences. Dialogue: 0,0:24:25.51,0:24:29.36,Default,,0000,0000,0000,,But sometimes they weren't... Dialogue: 0,0:24:30.37,0:24:33.82,Default,,0000,0000,0000,,actually, they weren't correct,\Nwith regards to the triple. Dialogue: 0,0:24:33.82,0:24:39.42,Default,,0000,0000,0000,,And I was curious if you had any ways\Nof doing validity checks of this site. Dialogue: 0,0:24:39.42,0:24:43.04,Default,,0000,0000,0000,,Sometimes the triple \Nis "subject, predicate, object," Dialogue: 0,0:24:43.04,0:24:46.11,Default,,0000,0000,0000,,but the language model says, Dialogue: 0,0:24:46.11,0:24:48.56,Default,,0000,0000,0000,,"Okay, this object is very rare, Dialogue: 0,0:24:48.56,0:24:51.74,Default,,0000,0000,0000,,I'm going to say you are born in San Jose, Dialogue: 0,0:24:51.74,0:24:55.06,Default,,0000,0000,0000,,instead of San Francisco or vice versa." Dialogue: 0,0:24:55.06,0:24:58.88,Default,,0000,0000,0000,,And I was curious\Nif you had come across this? Dialogue: 0,0:24:58.88,0:25:01.51,Default,,0000,0000,0000,,So that's what we call hallucinations. Dialogue: 0,0:25:01.51,0:25:05.08,Default,,0000,0000,0000,,The idea that \Nthere's something in a sentence Dialogue: 0,0:25:05.08,0:25:07.69,Default,,0000,0000,0000,,that wasn't in the original triple\Nand the data. Dialogue: 0,0:25:08.40,0:25:11.35,Default,,0000,0000,0000,,What we do-- \Nso we don't do anything about it, Dialogue: 0,0:25:11.35,0:25:13.91,Default,,0000,0000,0000,,we just also realized\Nthat that's happening. Dialogue: 0,0:25:13.91,0:25:15.91,Default,,0000,0000,0000,,It's even more happening\Nfor the low-resource, Dialogue: 0,0:25:15.91,0:25:19.73,Default,,0000,0000,0000,,because we work across domains,\Nso we are domain independently generating. Dialogue: 0,0:25:19.73,0:25:24.67,Default,,0000,0000,0000,,Traditional energy work \Nis always biography domain, usually. Dialogue: 0,0:25:24.67,0:25:26.62,Default,,0000,0000,0000,,So that happens a lot Dialogue: 0,0:25:26.62,0:25:29.51,Default,,0000,0000,0000,,because we just have little training data \Non the low-resource languages. Dialogue: 0,0:25:30.40,0:25:32.80,Default,,0000,0000,0000,,We have a few ideas. Dialogue: 0,0:25:32.80,0:25:36.84,Default,,0000,0000,0000,,It's one of the million topics, \NI'm supposed to work on at the moment. Dialogue: 0,0:25:38.85,0:25:42.55,Default,,0000,0000,0000,,One of them is to use \Nentity linking and relation extraction, Dialogue: 0,0:25:42.55,0:25:44.44,Default,,0000,0000,0000,,to align what we generate Dialogue: 0,0:25:44.44,0:25:46.64,Default,,0000,0000,0000,,with the triples\Nwe inputted in the first place, Dialogue: 0,0:25:46.64,0:25:50.75,Default,,0000,0000,0000,,to see if it's off or the network \Ngenerates information it shouldn't have Dialogue: 0,0:25:50.75,0:25:54.09,Default,,0000,0000,0000,,or it cannot know about, basically. Dialogue: 0,0:25:54.09,0:25:58.68,Default,,0000,0000,0000,,That's also all I can say about this\Nbecause now time is over. Dialogue: 0,0:25:58.68,0:26:01.48,Default,,0000,0000,0000,,(person 5) I'd love to talk offline\Nabout this, if you have time. Dialogue: 0,0:26:01.48,0:26:03.26,Default,,0000,0000,0000,,Yeah, absolutely, let's chat about it. Dialogue: 0,0:26:03.26,0:26:05.14,Default,,0000,0000,0000,,Thank you so much,\Neveryone, it was lovely. Dialogue: 0,0:26:05.14,0:26:06.60,Default,,0000,0000,0000,,(moderator) Thank you, Lucie. Dialogue: 0,0:26:06.60,0:26:08.61,Default,,0000,0000,0000,,(applause)