"normal" field for terms no longer exists in v13?

[scpedicini]

I used to use v11 heavily in a project a long time ago, and have been recently going through and updating it to use it with v13 compromise but it's missing some older properties/fields. I used to break apart each "compound verb" using v.list[0].terms and iterating through each term but the "normal" field is missing. Has "normal" been renamed? I noticed other fields 'reduced' and 'cleaned'. Can I substitute normal for one of these?

Also, I remember that 'silent_term' was related to contractions, but can't find mention of that field either.

[spencermountain]

hey @scpedicini yeah the internal api changes and will change again in v14. You should be able to break apart a match into individual terms with .terms() and get request any metadata, such as normalized, or implict words, with .json()
cheers

[spencermountain]

no prob. This is something I've struggled with too.

sometimes I do .clone().nouns().toSingular()

but there's also .cache():
https://runkit.com/spencermountain/60ef32911f1137001a28fd0c
That's how i'd do it. It's not perfect.
may be worth trying either, to see if it's faster to clone or run the cache.

i thought .json({terms:{root:true}}) would have worked, but maybe it's a bug.

once you've run the cache, you should be able to do

doc.match('~king~') //matches kings

agreed that this is all awkward right now, and I'd love to clean it up.

[spencermountain]

yeah, sometimes I think it's easier to just loop through json, and have people do custom stuff, but then you always end up having to re-connect back to the original document, like carry around indexes or offsets or something, which is hell.

it's frustrating that a compromise match result doesn't play as nicely as json - and feels so frameworky. I thought adding .map and .foreach would help with that, but it doesn't really feel like it.

[scpedicini]

thanks, yeah I get the frustation, I worked on a fairly substantial library where we had four or five different means of exposing the same data, but it just ended up getting really confusing and half the time you'd end up mixing the different accessor systems together which would get even more dicey.

Part of my problem is that I would guess that my use-case is somewhat unusual for compromise, in so far as I want to extract a noun(X) value and then use the tags on each term within the noun and reduce each term to a singular form, for example compromise would consider this the king's brother's dog's toys as a single noun, and from there I had formerly in v11 reduced each term using the list[0] and 'root' form that was automagically included.

But I think your suggestion of using toSingular() might solve this for me:
nlp("the king's brother's dog's toys").nouns().clone().nouns().toSingular().json()[0]
Unfortunately, it looks like it strips out original casing. Is that intentional?

nlp("Those are Great Danes").nouns(0).json()[0]
Each term "text" field contain original capitalized case "Great" and "Danes".

nlp("Those are Great Danes").nouns(0).clone().nouns(0).toSingular().json()[0]
Terms[0] => "Great"
Terms[1] => "dane" ; has been lowercased

[spencermountain]

that looks like a bug, please file it
thanks

[scpedicini]

Will do~
#855