Retrieving "RIGHT_ID" from tokens in DependencyMatcher

[felipemarineli]

Hi,

I am trying to retrieve the "RIGHT_ID" attribute for each token in a match using the DependencyMatcher. I could not figure out a way to do this. From the code snippet below, I would like to be able to output anchor and author. Is that possible?

    pattern = [
        # anchor token: founded
        {
            "RIGHT_ID": "anchor",
            "RIGHT_ATTRS": {"TEXT": {"REGEX": "[Bb]y"}}
        },
        {
            "LEFT_ID": "anchor",
            'REL_OP': ".*",
            "RIGHT_ID": "author",
            "RIGHT_ATTRS": {"ENT_TYPE": "PERSON"}
        }
    ]

    dependency_matcher.add("AuthorName", [pattern])

    for node in data:
        # apply NLP
        doc = nlp(input)  # Text by John Doe
        # apply RULER
        ruler_doc = ruler(doc)  # Text by John Doe
        # apply MATCHER on `doc` after RULER
        matches = matcher(ruler_doc)  # [(16787935407549884254, [1, 2])]
        ----- > print out RIGHT_ID of each match

Thank you in advance!

[adrianeboyd]

The list of matched tokens should correspond 1-to-1 to the pattern dicts, so [1, 2] means that the token with index 1 matched the first dict in the pattern and token with index 2 matched the second dict.

[felipemarineli]

Thank you for your reply!

I did manage to retrieve the RIGHT_ID in most cases by retrieving the original pattern dictionaries and using the list of matched tokens.

However, I do not seem to be able to retrieve the RIGHT_ID of the anchor token. Thus, in the example above, I can get author, but not anchor. This is fine for most cases, but in some cases, I am interested in the first token of the input Doc. For instance, if the input would be something like John Doe, Correspondent, I am interested in the correct matching of the whole thing (not any PERSON entity should be an author, but only the ones followed by "Correspondent"), but what I need to retrieve is the author bit (i.e. the RIGHT_ID of the anchor token).

Am I missing something?

[adrianeboyd]

Can you give a complete example that shows what you're trying to do? It's fine to use either manual annotation on the doc or a trained pipeline like en_core_web_sm (just specify which version so we can get the exact same annotation when we run the example on our end).

[felipemarineli]

Hi Ariane, thank you once again for your reply. I found the reason why I was not able to retrieve the RIGHT_ID of the anchor token, and it was not related to spaCy. I did manage now to do what I intended. Thank you :)