4 ms·

The History of Text Generation

57 points by polm23 8y ago

enriquto 8y ago
Warning: if you want to read this text without javascript you have to disable css, otherwise all the content is hidden. So, it has come to that. This is the new web.
- dfrage 8y ago
  Thanks for the tip! I use uMatrix with aggressive defaults including no Javascript, somehow this trick didn't occur to me. Although my general rule is that if Javascript is required to read a site's content, it's probably not worth it. With the storied NoScript being released for Chrome (https://news.ycombinator.com/item?id=19650731 https://news.ycombinator.com/item?id=19650731), I suspect I'll have more and more company as the new web gets more abusive and dangerous.
  - enriquto 8y ago
    I am so in awe at this site! It is the first actual example I have witnessed of non-monotonic progressive enhancement. You start with a text-only browser and the site is OK; so far so good. Then you keep adding features to the browser and the site progressively disappears and re-appears again.
- ChrisSD 8y ago
  Or right-click inspect element and delete the div overlay. The page is already fully loaded but they seem to add an artificial delay for no reason. However that is the modern way (see also: amp). I guess browsers are just too gosh darn fast nowadays.
- spc476 8y ago
  If you are on Firefox and have Javascript disabled, you can select the menu item “View → Page Style → No Style” and see the content.
aantix 8y ago
Are there any natural text generation libraries that can generate commentary on time series data? E.g. blood glucose levels remained steady for the first two months then rose sharply in the past week.
- gattilorenz 8y ago
  Well, yes and no. Yes in the sense that there are libraries that can generate text (e.g. SimpleNLG), but you need to write code that does the analysis and chooses the content to show. The "modern" way would be to train a seq2seq model. But for that you need training data, and it's harder to control the correctness of the results. For a start, you can look at the slides from this summer school https://nlgsummer.github.io https://nlgsummer.github.io and read up on the BabyTalk project (http://www.nlg-wiki.org/systems/BabyTalk http://www.nlg-wiki.org/systems/BabyTalk). SimpleNLG is indeed simple to use, but maybe for your task a template-based system (e.g. https://github.com/TallChris91/PASS https://github.com/TallChris91/PASS) is enough. TLDR: you either need to write code for managing the time series, or need a large enough dataset to train on. Or rely on commercial services such as those offered by Arria
gattilorenz 8y ago
Wow, basically grammar-based generation has been a thing for quite a while now. I'll have to print one of these circles and bring it to my NLG group to show...

57 points by polm23 8y ago

Warning: if you want to read this text without javascript you have to disable css, otherwise all the content is hidden. So, it has come to that. This is the new web.

dfrage 8y ago
Thanks for the tip! I use uMatrix with aggressive defaults including no Javascript, somehow this trick didn't occur to me. Although my general rule is that if Javascript is required to read a site's content, it's probably not worth it. With the storied NoScript being released for Chrome (https://news.ycombinator.com/item?id=19650731 https://news.ycombinator.com/item?id=19650731), I suspect I'll have more and more company as the new web gets more abusive and dangerous.
- enriquto 8y ago
  I am so in awe at this site! It is the first actual example I have witnessed of non-monotonic progressive enhancement. You start with a text-only browser and the site is OK; so far so good. Then you keep adding features to the browser and the site progressively disappears and re-appears again.
ChrisSD 8y ago
Or right-click inspect element and delete the div overlay. The page is already fully loaded but they seem to add an artificial delay for no reason. However that is the modern way (see also: amp). I guess browsers are just too gosh darn fast nowadays.
spc476 8y ago
If you are on Firefox and have Javascript disabled, you can select the menu item “View → Page Style → No Style” and see the content.

aantix 8y ago

Are there any natural text generation libraries that can generate commentary on time series data? E.g. blood glucose levels remained steady for the first two months then rose sharply in the past week.

gattilorenz 8y ago
Well, yes and no. Yes in the sense that there are libraries that can generate text (e.g. SimpleNLG), but you need to write code that does the analysis and chooses the content to show. The "modern" way would be to train a seq2seq model. But for that you need training data, and it's harder to control the correctness of the results. For a start, you can look at the slides from this summer school https://nlgsummer.github.io https://nlgsummer.github.io and read up on the BabyTalk project (http://www.nlg-wiki.org/systems/BabyTalk http://www.nlg-wiki.org/systems/BabyTalk). SimpleNLG is indeed simple to use, but maybe for your task a template-based system (e.g. https://github.com/TallChris91/PASS https://github.com/TallChris91/PASS) is enough. TLDR: you either need to write code for managing the time series, or need a large enough dataset to train on. Or rely on commercial services such as those offered by Arria

gattilorenz 8y ago

Wow, basically grammar-based generation has been a thing for quite a while now. I'll have to print one of these circles and bring it to my NLG group to show...