mneb/genia2013

GENIA 2013 (mneb format) — joint NER + nested event extraction

The BioNLP Shared Task 2013 GE corpus (Kim et al., 2013) converted into the mneb joint entities + json_structures format. It is a biomedical entity and end-to-end event extraction dataset over PMC Open Access full-text articles.

This dataset keeps event-as-argument nesting. Roughly a third of GENIA's argument links point at another event rather than an entity — regulation events take other events as their Theme or Cause — and prior conversions (including TextEE's) drop them. Here they are kept without leaving the flat mneb record shape: every argument is a plain {role,text,start,end} span, and an argument that points at an event carries that event's trigger span. See Event arguments. GENIA 2013 has the deepest nesting of the three BioNLP event corpora, reaching depth 6.

Char offsets are character-based and end-exclusive (input[start:end] == text). One record = one source .txt file. GENIA 2013 is all full-text, so the 471 records are the sections of 10 train + 10 devel papers; sections are not merged per paper.

Splits

Split	Records	With events	Events
train	222	149	2,673
validation	249	157	2,944
Total	471	306	5,617

There is no labelled test split. The available release ships .txt/.a1 for the blind test documents but no .a2, i.e. no public gold events. Treating these inputs as empty output would create false negatives. Note that this differs from TextEE, which merges sections into 20 papers and makes five random re-splits, so no number reported on a TextEE split is directly comparable to this one.

Supported tasks and retained layers

• NER: protein and generic entity mentions in output.entities.
• Event detection (ED): event triggers in output.json_structures.
• Event argument extraction (EAE) and end-to-end event extraction (E2E): trigger and argument spans, including span-linked event arguments.
• Joint NER + EE: both layers occur in the same record and use the same character offsets.

Coreference is on hold. The 222 raw Anaphora mentions and the source Coreference relations are deliberately excluded together; exposing anaphor spans without their links would misrepresent that annotation layer. No emitted event argument directly targets an Anaphora, so this choice does not alter the retained EE layer. Equivalence, attributes and normalisations are also omitted. This release therefore does not claim RE/coreference, negation or speculation.

Entity layer — 2 types, 8,144 mentions

Entity type	Train	Validation	Total
entity	121	314	435
protein	3,571	4,138	7,709
Total	3,692	4,452	8,144

protein is the public form of .a1 label Protein; entity is the public form of the generic .a2 label Entity used for non-protein physical participants, sites and locations. The names are uniformly lower-case with spaces.

Record format

{
  "input":  "<document text>",
  "output": {
    "entities": {"protein": [<span>, ...], "entity": [<span>, ...]},
    "json_structures": {"<event type>": [<event>, ...]}
  },
  "schema": {"entities": [<entity definition>, ...],
             "json_structures": [<event definition>, ...]}
}

The active data/train.jsonl and data/validation.jsonl use lower-case, space-separated event labels (for example gene expression), preserving the convention introduced by yangwang825. Compatibility files prefixed genia2013_ retain the official event strings (for example Gene_expression). NER labels are normalized in both views.

An event is {"trigger": {"text","start","end"}, "arguments": [<arg>, ...]}. There is nothing else: no type field on the event (its type is the json_structures key), and no object nested inside an argument.

Event arguments

Every argument has the same four keys, whether it points at an entity or at another event:

{"role": "Theme", "text": "IL-4",            "start": 200, "end": 204}
{"role": "Theme", "text": "mRNA expression", "start": 217, "end": 232}

The first is an entity mention. The second is an event link: (217, 232) is the trigger span of a transcription event, which is listed at the top level of the same record. This is how mneb expresses links generally — repeat the span, no ids (cf. mneb/bc5cdr, whose relation head/tail repeat the entity spans).

Because a linked child must be reachable, every event appears at the top level, not only the roots. To resolve links:

def resolve(js):
    """Index every event by its trigger span, then read arguments as links where they match."""
    by_span = {}
    for etype, evs in js.items():
        for ev in evs:
            by_span.setdefault((ev["trigger"]["start"], ev["trigger"]["end"]), []).append((etype, ev))
    for etype, evs in js.items():
        for ev in evs:
            for a in ev["arguments"]:
                target = by_span.get((a["start"], a["end"]))      # None => entity mention
                yield etype, ev, a, target

The deepest chain in the corpus, from PMC-2674207-16-Results — "This effect was concentration- and time-dependent… and was associated with marked reductions in anti-CD3/CD28-stimulated IL-4 and IL-5 mRNA expression" — is now six separate top-level events, chained by Theme arguments that land on the next one's trigger span:

regulation("dependent")            --Theme--> span of "This effect"
  regulation("This effect")         --Theme--> span of "associated"
    regulation("associated")         --Theme--> span of "reductions"
      negative regulation("reductions") --Theme--> span of "stimulated"
        positive regulation("stimulated") --Theme--> span of "mRNA expression"
          transcription("mRNA expression") --Theme--> "IL-4"   (entity)

How faithful the span links are

• Telling a link from an entity mention: 0 of the 5,679 entity-valued arguments sit on a span that is also a trigger, so the test "this argument's span matches a trigger span" has no false positives here.
• Telling which event a link points at: 1,817 of the 2,478 links (73.3%) match exactly one event of the right type. The other 661 (26.7%) land on a trigger span shared by several same-type events, and the span cannot disambiguate them.
• Consequently 399 of the 6,016 raw E lines (6.6%) come out byte-identical to another entry of the same type and are collapsed, leaving 5,617 events. Those are exactly the parents that differed only in an unresolvable choice of child; keeping both copies would double-count in any set-based metric.

Everything else round-trips: the offset invariant holds on every span, and the set of emitted (type, trigger, role/span) signatures equals the same set computed straight off the raw standoff, for every document.

Statistics

• 13 event types, 7 role types (role strings kept verbatim, so Theme2/Site2 are not collapsed into their base role).
• 8,157 raw argument links = 5,679 entity-valued + 2,478 event-valued (30.4% of all argument links are event-to-event). After the collapse above the files hold 7,443 argument instances = 5,474 entity spans + 1,969 span links.
• 2,117 events (35.2%) take at least one event argument.
• Raw nesting depth histogram {1: 3899, 2: 1789, 3: 297, 4: 26, 5: 3, 6: 2} — max depth 6.
• Nesting is driven entirely by the three regulation types; no other event type ever takes an event argument, and only Theme and Cause are ever event-linked.
• The four post-translational-modification types (Protein_modification, Ubiquitination, Deacetylation, Acetylation) are new relative to GENIA 2011 and are very rare (≤9 each).

Full type/role inventory and nesting patterns: genia2013_label_summary.md. Browsable rendering: genia2013_vis.html (open directly; data embedded, no server needed).

How this was derived

Built from the original-data/{train,devel} standoff files:

1. Each document's .a1 (Protein entities) and .a2 (event triggers, Entity/Anaphora spans and E event lines) are parsed into a single text-bound annotation map; their id spaces do not collide.
2. Protein and Entity annotations become NER mentions under normalized public labels. Anaphora is omitted while its associated coreference task is on hold.
3. Each E line becomes an event grouped under its own type; Role:T… arguments become the entity's span and Role:E… arguments become the child event's trigger span. Both come out in the same {role,text,start,end} shape.
4. Every event is listed at the top level, so a linked child is always resolvable. Entries that are byte-identical under one event type are then collapsed.
5. Relation (R, including Coreference), equivalence (*), attribute (A/M: Negation, Speculation) and normalisation (N) lines are not carried over. The converter and compatibility files preserve official event-type strings; the active files normalize only event-type labels. Roles remain verbatim in both views.

The raw standoff gives labels but no prose label descriptions. The descriptions embedded in the full schema and published in schema.json are concise mneb-authored paraphrases based on the official task definition and paper; they are not source annotations or quotations.

Verification built into the converter: every emitted span is re-checked against the source text (input[start:end] == text, 0 failures); the set of emitted (type, trigger, role/span) signatures is compared against the same set computed straight off the raw standoff and is equal for every document; and a link audit reports, for every event-valued argument, whether its child is uniquely identifiable from the span (the numbers quoted above).

Source provenance. The raw data comes from the community mirror openbiocorpora/bionlp-st-2013-ge, which preserves the original .txt/.a1/.a2 files under original-data/. The official 2013.bionlp-st.org rev3 tarball links are no longer served.

Licence and terms of use

• Annotations are licensed under the Creative Commons Attribution 3.0 Unported licence (CC BY 3.0). Copyright in the annotations belongs to the Database Center for Life Science (DBCLS).
• Full texts come from the PMC Open Access Subset. Each article is protected by copyright but made available under a Creative Commons or similar licence; use and distribution of the texts is subject to the terms applying to each individual publication.

Citation

@inproceedings{Kim13genia2013,
  author    = {Jin-Dong Kim and Yue Wang and Yamamoto Yasunori},
  title     = {The Genia Event Extraction Shared Task, 2013 Edition - Overview},
  booktitle = {Proceedings of the BioNLP Shared Task 2013 Workshop},
  year      = {2013}
}