xref extract: ValueError on capitalized "VerbNet" lex/rolelink resource (only lowercase handled)

[aaronstevenwhite]

Summary

glazing xref extract aborts with:

✗ Error: Unknown lexlink resource type: VerbNet

(and the symmetric Unknown rolelink resource type: VerbNet for role-level links).

Root cause

In src/glazing/references/extractor.py, _index_propbank_mappings normalizes FrameNet's capitalization variants but matches VerbNet only in lowercase:

# lexlinks (~line 302)
if lexlink.resource == "verbnet":                 # lowercase only
    target_dataset = "verbnet"
elif lexlink.resource in ["FrameNet", "Framenet"]:
    target_dataset = "framenet"
else:
    raise ValueError(f"Unknown lexlink resource type: {lexlink.resource}")

# rolelinks (~line 335) — same asymmetry
if rolelink.resource == "verbnet":
    ...
elif rolelink.resource in ["FrameNet", "Framenet", "framenet"]:
    ...
else:
    raise ValueError(f"Unknown rolelink resource type: {rolelink.resource}")

But the PropBank frame files overwhelmingly use the capitalized form. In propbank-frames/frames/*.xml:

39821 resource="VerbNet"
11322 resource="FrameNet"
11211 resource="AMR"
   56053 resource="PropBank"
    4 resource="Framenet"

And src/glazing/types.py explicitly lists "VerbNet" (capitalized) as a valid ResourceType ("Variant capitalization found in some files"). So a VerbNet lex/rolelink is well-typed data that the extractor nonetheless rejects, falling through to the else and raising.

Reproduction

glazing init
glazing xref extract
# ValueError: Unknown lexlink resource type: VerbNet

Impact

xref extract cannot complete on the standard SemLink-annotated PropBank frames, so no cross-reference index is produced. This blocks any downstream consumer of references/SemLink mappings.

Suggested fix

Normalize VerbNet the way FrameNet is normalized (both branches), e.g.:

if lexlink.resource in ("verbnet", "VerbNet", "Verbnet"):
    target_dataset = "verbnet"

A case-insensitive comparison (or reusing a single normalization helper over the ResourceType variants declared in types.py) would also cover the AMR resource (11,211 occurrences), which would otherwise raise here as well.

Found on glazing==0.3.0.