Comment by worldsayshi | Hacker News Reader

Comment by worldsayshi | Hacker News Reader

Parent Full thread

In brief: Let's assume that we can build a formal DSL for expressing the discussion that we want to have. Then we can build a parser and natural language "code completer" that looks at your sentences as you type, checks partial sentences against a search index (ideally a well structured one) to lookup their 'type' or does pos-tagging some other way. Then it tries to match the tagged words and their order against the DSL. This matcher should be able to suggest syntactically valid natural language sentences that the user then selects from - as they type.

The point of using a search engine (like elastic) is that you can easily match against n-gram/fuzzy versions of the words. So you can write in a sloppy way and still have it be parse-able. (But that's standard code completion I guess.)

Possible, but how do you explain this to the less technically-minded people?

Not sure what you refer to when you are saying "this". The completion part or to explain the domain (of possibilities)?

Ideally, if you design it right you shouldn't need to explain the text completion. It adds value by allowing faster typing.

So ideally, you shouldn't need to motivate the feature further. It should pay for itself UX wise. This very much depends on you being able to match your DSL to the domain being discussed. If the domain is politics it is very hard but if the domain is discussion meta, it might be easier.

Another follow up issue is that you want to do more with the structured text. You want it to be interactive, and have effect on the rest of the UI. You want to reflect the semantics of what you're writing. For example, perhaps you can pick out data points from the DSL interpretation and pull it into a common visualization.

Compare with tagging people on Facebook. People do this sort of naturally. This could work in a similar way. Instead of showing suggestions when typing "Bob Smi" it would be suggestions when typing things like "child mortality in ". And the effect of the complete sentence could be to attach relevant data.

Reply on news.ycombinator.com