[14:09:04] Hi everyone! [14:09:40] I've been following this project from a (far-ish) distance for some time, super excited about it! [15:47:32] hello, did anyone read / come by / is familiar with https://direct.mit.edu/books/book/5042/Linguistics-for-the-Age-of-AI [15:47:45] it is open access [17:44:34] Welcome Michał, great to see you here! [17:45:47] I looked through the table of contents, and the whole book seems to be about parsing and reasoning? [18:02:01] I am not the writer of the book if you are talking to me [18:02:38] it is about how to build application support by knowledge toward AI-complete systems [18:04:54] vrandecic: what are you replying to, I backlogged I see no book about parsing or reasoning. [18:05:47] At this point in the book I mentioned, they use Standford corenlp for parsing (componenent and stuff dependency, POS, possibly other stuff) [18:06:19] they describe reasoning but not in a great technical details. [18:07:08] they use generic lexicon [18:07:19] afaiu they have domain ontologies for each applications [18:07:45] I guess they do backward / foward chaining. [18:09:25] It unclear whether it intersect with the goals of AbstractWikipedia *and* WikiFunctions [18:09:57] but clearly, the undescore that some knowledge engineering should be repurposed to rely on the technics they describe [18:10:19] they write they have base lexicon of 30K senses [18:10:48] I guess you should contact the authors and see how to interop the projects [18:12:41] I am nobody, like past xp has prooved, they will not trust me, and I have no power of any sort around mediawiki projects to repurpose the vast amount of effort and energy available to it [18:14:21] I will go around the wiki functions project, without considering the broken existing wikidata sofware codebase and infrastructure, and try to forgo wasted scares energy taken from our shared environment. [18:14:58] I am available for questions about the book, tho. [18:15:41] (re wiki functions, it is unclear to me what progress in the roadmap was made since the initial announcement in 2020) [18:15:55] I believe this is why Denny asked the question he asked, and so clarifying that would surely help (re @wmtelegram_bot: It unclear whether it intersect with the goals of AbstractWikipedia *and* WikiFunctions) [18:16:41] oh absolutely, I don't think any system for Abstract Wikipedia could be made without such reliance on described techniques (re @wmtelegram_bot: but clearly, the undescore that some knowledge engineering should be repurposed to rely on the technics they describe) [18:17:15] there's a lot of senses across languages among Wikidata lexemes, although there could certainly be more (re @wmtelegram_bot: they write they have base lexicon of 30K senses) [18:17:33] no need to be so self-deprecating :-) where did this line of thought come from? (re @wmtelegram_bot: I am nobody, like past xp has prooved, they will not trust me, and I have no power of any sort around mediawiki projects to repurpose the vast amount of effort and energy available to it) [18:18:38] ? not sure how you got to this point from Denny's original question (re @wmtelegram_bot: I will go around the wiki functions project, without considering the broken existing wikidata sofware codebase and infrastructure, and try to forgo wasted scares energy taken from our shared environment.) [18:21:01] I am not self deprecating, that is a statement from the book, 30K senses in the lexicon, I am too lazy to look throught the pdf the find it. [18:21:47] by "self-deprecating" I was referring to the message "I am nobody...I have no power..." [18:22:38] nevermind. [18:23:34] yes, because I tried to stire wikidata forward made a proposal to improve the situation with the code base and offered a path to scale with ease and it was refused, and dubbed impossible. [18:23:35] I did find the statement in the book that there are 30K senses for English; the overlap with Wikidata's English senses may not be that great, but there's definitely a lot of overlap with senses in other languages [18:24:17] yes, but properties are different compared to a regular or even wiki data lexicon [18:24:27] wiki data lexemes [18:24:48] by "Wikidata senses" I'm referring to the senses on an object such as L52273 [18:25:10] they use some kind of typed feature structures with holes to describe words, and some type of words such verbs, modifiers [18:25:56] I have slowly been developing a system to generate text with Wikidata items and lexemes (where these lexemes can be words or phrases) by the creation and manipulation of dependency grammar trees [18:26:49] whether wikidata has more senses or less, does not matter, they claim interop with existing system will be difficult. Surely they know about wiki projects. [18:26:57] It's of course a bit limited in what it can do at the moment (since I've only written some renderers for certain constructors, and even then for Bengali and sometimes German and Swedish), but the potential is there at the moment [18:27:29] what is the science or engineering background for such work? [18:28:20] do you mean what background I have that drives me to do this? [18:28:40] no, what makes you think or believe that is the good way forward ? [18:29:24] by "that" do you mean "making an NLG system in the first place", "using Wikidata as a basis for such a thing", "using dependency grammar trees", or something else? [18:29:31] yes [18:31:44] for "making an NLG system in the first place" see https://arxiv.org/abs/2004.04733; for the other things it comes down to 1) I've got a hunch, 2) I'm tired of the disfluent answers on the subject from the AW team, and 3) I'd like to not defer to the divine in terms of making progress [18:34:29] 1) good 2) hmm 3) awesome [18:34:35] re reading the paper a this time [18:37:07] the system I'm building, which resides at https://bitbucket.org/mmorshe2/ninai/ (and depends on the libraries residing at mmorshe2/tfsl/ and mmorshe2/udiron/), might benefit from attempting to write fus'ha renderers in addition to the bn/de/sv ones currently present [18:38:42] given that the languages attempted to be supported are all indo-european, there may definitely be places where adjustments will be necessary (it's early enough that performing the drastic isn't a problem) [18:39:14] *the three languages currently attempted to be supported [18:40:51] (also renderers for amazigh languages would be nice to have) [18:41:08] too bad you are not on irc.libera.chat [18:42:41] ah yes, I don't have an IRC client running all the time, but sometimes I do [18:42:54] so you know amazigh :D [18:43:19] I do not how to write or read any dialect amazigh, but I can speak :) [18:43:22] I know that there's a set of languages referred to with that name, but I do not know anything about the languages themselves [18:43:54] there is two alphabets: tifinag and an alphabet inspired from french / latin [18:44:13] I prefer tifinag, but people around me say latin is more modern [19:23:36] (if all messages since 15:47 UTC in this chat could be wiped, that would be great I'm sure) [19:27:39] (as those exploring the logs between then and now will find nothing of substance within)