The main dependency conflicts with 2nd dependency at ID={i},

Error message

The main dependency conflicts with 2nd dependency at ID={i}, which means joint mode might not be suitable. The sample is
{sample_str}

What it means

When merging main and second-order dependencies for SemEval-15 joint parsing, if the 2nd-best graph already has a different label for the head chosen by the main parser, the labels conflict and joint mode may be inconsistent; the full CoNLL sample is printed in the warning.

Source

Thrown at hanlp/datasets/parsing/semeval15.py:56

    return sample


def append_bos_to_form_pos(sample, pos_key='CPOS'):
    sample['token'] = [ROOT] + sample['FORM']
    if pos_key in sample:
        sample['pos'] = [ROOT] + sample[pos_key]
    return sample


def merge_head_deprel_with_2nd(sample: dict):
    if 'arc' in sample:
        arc_2nd = sample['arc_2nd']
        rel_2nd = sample['rel_2nd']
        for i, (arc, rel) in enumerate(zip(sample['arc'], sample['rel'])):
            if i:
                if arc_2nd[i][arc] and rel_2nd[i][arc] != rel:
                    sample_str = CoNLLSentence.from_dict(sample, conllu=True).to_markdown()
                    warnings.warn(f'The main dependency conflicts with 2nd dependency at ID={i}, ' \
                                  'which means joint mode might not be suitable. ' \
                                  f'The sample is\n{sample_str}')
                arc_2nd[i][arc] = True
                rel_2nd[i][arc] = rel
    return sample

View on GitHub (pinned to ddb1299bdd)

Solutions

Check the printed sample at the flagged ID to see which deprel is intended
If labels disagree, prefer non-joint (separate) decoding for that data or fix the source annotation
Ignore if occasional disagreement is acceptable for your training setup

Defensive patterns

Strategy: try-catch

Validate before calling

for i, (arc, rel) in enumerate(zip(sample['arc'], sample['rel'])):
    if i and sample['arc_2nd'][i][arc] and sample['rel_2nd'][i][arc] != rel:
        conflicts.append(i)  # inspect before joint mode

Try / catch

with warnings.catch_warnings():
    warnings.filterwarnings('ignore', message='.*joint mode might not be suitable.*')
    merged = merge_head_deprel_with_2nd(sample)

Prevention

Audit SemEval-15 samples for head/label conflicts before joint training
Fall back to non-joint parsing when conflicts are frequent

When it happens

Trigger: Calling merge_head_deprel_with_2nd on a SemEval-15 sample where arc_2nd[i][arc] is True but rel_2nd[i][arc] != rel — i.e. both graphs pick the same head with different deprel labels.

Common situations: Preparing SemEval-15 data for joint dependency parsing; mismatched main/2nd annotations in the corpus or preprocessed files.

Related errors

AI-assisted analysis of hankcs/HanLP@ddb1299bdd (2026-08-27). Data as JSON: /api/errors/159ba71814201716. Report an issue: GitHub.