Conjuncts are not identified completely in spaCy

[Xinxin-Harbsafe2]

Hi,

I have a problem while using the .conjuncts, i.e., not all conjuncts are identified.

As described in the following link, anyone knows the reason? Thanks!

[polm]

Your question doesn't have any concrete examples. Can you give a specific example where you're not seeing the output you want, with sample code?

Based on your description I'm not sure if you're seeing a bug or not.

[polm]

Looks perfectly fine to me?

import spacy

nlp = spacy.load("en_core_web_md")

text = 'software program that performs specific functions initiated by a user command or a process event and that can be executed without access to system control, monitoring, or administrative privileges'

doc = nlp(text)

for chunk in doc.noun_chunks:
    print("=====")
    print(chunk)
    print(chunk.conjuncts)

Output:

=====
software program
()
=====
specific functions
()
=====
a user command
(event,)
=====
a process event
(command,)
=====
access
()
=====
system control
(monitoring, privileges)
=====
monitoring
(control, privileges)
=====
administrative privileges
(control, monitoring)

Not sure why you're using prev_end instead of just using the chunk spans, is there a particular reason for that?

[Xinxin-Harbsafe2]

Thanks.

why...using prev_end instead of just using the chunk spans...
Yes, I have a reason, because I want to split the sentence at the end of each chunk, but not only get the chunks.

I am wondering, why it does not identify the a user command (event,) and a process event(command,) while using prev_end?

Have you tried using the prev_end?
Thanks again!

[polm]

OK, here's code using prev_end. I also added the output of the root of each span so it's easier to see what's going on.

import spacy

nlp = spacy.load("en_core_web_md")

text = 'software program that performs specific functions initiated by a user command or a process event and that can be executed without access to system control, monitoring, or administrative privileges'

doc = nlp(text)

prev_end = 0
for chunk in doc.noun_chunks:
    print("=====")
    span = doc[prev_end:chunk.end]
    print(span, f"(root: {span.root})")
    print(span.conjuncts)
    prev_end = chunk.end

And here's the output:

=====
software program (root: program)
()
=====
that performs specific functions (root: performs)
(executed,)
=====
initiated by a user command (root: initiated)
()
=====
or a process event (root: or)
()
=====
and that can be executed without access (root: and)
()
=====
to system control (root: to)
()
=====
, monitoring (root: monitoring)
(control, privileges)
=====
, or administrative privileges (root: or)
()

If you look at the printed root in these spans, you can see that the head of the noun chunk is often not the root. Since conjuncts only returns conjuncts of the root of the span, it won't work the same as the test code using noun_chunks instead of prev_end. For example, in "initiated by a user command", the root is "initiated", so only conjuncts of that verb, not conjuncts of the noun "command", will be returned.

Does that make sense?

[Xinxin-Harbsafe2]

Thanks a lot! I should not use the conj_ = span.conjuncts, I should use the conj_ = chunk.conjuncts then. Thanks again!

[polm]

Note that the chunk is also a span - you just need to use a span where span.root is the token you want to get conjuncts for.