News for hackers. Stuff that matters.
Re:Natural language generation: The commercial state of the ... by is_true on Monday January 11, 2021 @08:06AM (#25728357)
I think that most use cases for NLG are unethical.

I provide raw data for some news services and that data could be shown as tables or simple dashboards that are faster to read for users, but my customers insist in generating as much text as possible from, sometimes, 2 values.

The incentives are totally misaligned, news services try to get as much time from it's readers as possible. It's quite depressing.

Re:Natural language generation: The commercial state of the ... by Saad_M on Monday January 11, 2021 @11:08AM (#25730799)
I very much disagree. There are use case such as weather reporting, electronic medical report summarisation, etc. That make more sense as a textual representation than bundles of graphs and tables. In fact textual summarisation has been shown to lead to better decision making[1].

Good data-to-text NLG applications not only summarise data but they also can provides insight into causal relations of why events occurred by leveraging domain knowledge.

[1] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2656014/

Re:Natural language generation: The commercial state of the ... by is_true on Monday January 11, 2021 @11:46AM (#25731511)
I agree with you Saad. I worded that comment poorly. There are some use cases that are gonna be used to keep people reading a lot of algorithmical gibberish for no reason other than increasing revenue, for example sports articles.

I think NLG is really interesting, the problem is that there are incentives to create long form content that doesn't add any value to readers.

Re:Natural language generation: The commercial state of the ... by coding123 on Monday January 11, 2021 @10:23AM (#25730095)
I tend to agree. While technologically its a marvel. But I do have the feeling the Robocallers and NLG customers are hanging out in the same place.
Re:Natural language generation: The commercial state of the ... by gxqoz on Monday January 11, 2021 @11:14AM (#25730915)
I did some experiments last summer trying to get GPT-3 to author quizbowl questions, a type of trivia question. As expected, GPT-3's loose relationship with the truth made these results not acceptable. But a system that really could generate new questions from existing facts would be immensely useful in this domain.
Re:Natural language generation: The commercial state of the ... by polm23 on Monday January 11, 2021 @11:35AM (#25731311)
I'm not surprised GPT is awful at that, but there are good ways to do it. Look for papers on "factoid question generation".
Re:Natural language generation: The commercial state of the ... by gxqoz on Monday January 11, 2021 @01:36PM (#25733673)
Quizbowl questions have a particular structure that makes them generally not work with these other approaches. For instance, it needs to generate paragraph-long questions with the clues that go from the most to least difficult. This is hard skill for a human to learn, much less a machine.
Re:Natural language generation: The commercial state of the ... by polm23 on Tuesday January 12, 2021 @02:09AM (#25743404)
Ah, that does sound difficult.
Re:Natural language generation: The commercial state of the ... by mkl95 on Monday January 11, 2021 @09:06AM (#25728979)
Usage of these algorithms and heuristics is starting to become quite obvious in Spanish sports journalism. I stumble into articles that are basically AI gibberish generated from a short quote every other day.
Re:Natural language generation: The commercial state of the ... by tyingq on Monday January 11, 2021 @10:34AM (#25730247)
Google is probably going to struggle with SEO content/link mills that now produce something better than spun gibberish.
Re:Natural language generation: The commercial state of the ... by lumost on Monday January 11, 2021 @11:07AM (#25730782)
I'd expect the only rational decision would be to ban the content mill domains from the crawler. Is there really that much value for the search customer in being directed to a generated page?
Re:Natural language generation: The commercial state of the ... by tyingq on Monday January 11, 2021 @11:42AM (#25731454)
That's the dilemma. NLG of sufficient quality[1] won't be easily detectable as machine generated.

[1] "Quality" here meaning how it looks to Google's crawler, not actual quality content.

Re:Natural language generation: The commercial state of the ... by lumost on Monday January 11, 2021 @02:57PM (#25735485)
domains auto-generating junk or derivative content should be detectable from both human annotation or automated means. If a site with low click through rates, traffic and extreme volumes of text appears, it's worth a re-examination by people.

In particular at search time, google has a vested interest in limiting both the number of links to a particular domain and the occurrence of content farm links when Wikipedia would suffice. Content farms are pretty detectable in any objective relevance annotation workflow.

Re:Natural language generation: The commercial state of the ... by tyingq on Monday January 11, 2021 @03:35PM (#25736256)
It wasn't that long ago that really poor quality auto generated content was clearly working for SEO purposes, so I'm not convinced Google is ahead of current state of the art.
Re:Natural language generation: The commercial state of the ... by is_true on Tuesday January 12, 2021 @01:21PM (#25751061)
It is working as of now.