Getting lemmas for all words in sentence

[moskaliukua]

I'm trying to process a sentence, to get base form for each word
for example: I am going to the store => I be go to the store
e.g. convert all verbs to infintive forms
I already tried a couple of options

import nlp from 'compromise';
const sentence = ' I am going to the store'
const doc = nlp(sentence);
doc.normalize({
      verbs: true,
}).out()

and
doc.verbs().toInfinitive().out()
produces the same result:
I go to the store
the other solution I tried is:

doc.terms().map((view) => {
          return view
            .normalize({
              verbs: true,
              plurals: true,
            })
            .out();
        }),

which produces this result:
[
'I',
'I be going to the store',
'I be go to the store',
'to',
'the',
'store'
]
and

doc.terms().map((view) => {
          const term = view.json();
          return nlp(term)
            .normalize({
              verbs: true,
            })
            .out();
        }),

is also not enough:
[ 'I ', 'am ', 'go', 'to ', 'the ', 'store' ]
I look carefully through the docs, it seems that I missing something. Any help would be appreciated.
Is there a better way to achieve what I'm trying to do?

[spencermountain]

hi @moskaliukua - i think this is probably what you're looking for -

let doc = nlp(' I am going to the store')
doc.compute('root')
doc.json()

https://runkit.com/spencermountain/66b243b9e0d0c900085ae794
you can see the documentation for root-words here
cheers

[moskaliukua]

thanks, that's what I'm looking for, thought is should be converted to be, but it shouldn't be hard to achieve it from this point

[hchiam]

Since your goal is "I am going to the store" => "I be go to the store", it tells me that each word needs to be processed, since "am going" and "am going to" can sometimes be considered as being derived from "to go" in some contexts.

And since you want "be" instead of "is", here are 2 alternatives: (i noticed that "am" and also "had" were otherwise not turning into their infinitives "be" and "have")

const sentence = 'I am going to the store';
const docs = sentence.split(' ').map(word => nlp(word));
docs.forEach(word => word.verbs().toInfinitive());
const output_1 = docs.map(doc => doc.text()).join(' ');
document.body.querySelector('#output_1').innerText = output_1;
// I be go to the store

or:

const sentence = 'I am going to the store';
const doc = nlp(sentence);
const output_2 = doc.terms().map(view => {
  const wordDoc = nlp(view.text());
  wordDoc.verbs().toInfinitive();
  return wordDoc.text();
}).join(' ');
document.body.querySelector('#output_2').innerText = output_2;
// I be go to the store

Here's a demo you edit live without login: https://codepen.io/hchiam/pen/NWZgORW?editors=1000

[moskaliukua]

hi @hchiam,
thanks for reply,
Tried all solutions, what is interesting:
the first solution(with multiple docs) is the slowest 15ms,
second one - 5ms, I also found similar one:

doc.match('#Verb').normalize({
      verbs: true,
    });

not sure if this is intentend to work this way
But solution with compute('root') is only 1ms. so I picked it up and just manually mapped is and had to infinitive forms