BERT for Event Expression Recognition

Overview

This model is a fine-tuned version of google-bert/bert-base-cased for event expression recognition.

The model performs token-level sequence labeling to identify event expressions in natural language text using BIO tagging.

Labels

• 0: O
• 1: B-E
• 2: I-E

where:

• O represents tokens outside an event expression.
• B-E represents the beginning of an event expression.
• I-E represents a token inside an event expression.

---

Task

Event Expression Recognition

Given an input text, the model identifies spans corresponding to event expressions.

The task is formulated as token classification using BIO sequence labeling.

---

Purpose

This model was developed as a supporting model for adversarial dataset construction rather than as a primary model for downstream event or temporal reasoning experiments.

Its purpose is to identify the locations of event expressions in the original input text before adversarial transformations are applied.

Event-based reasoning examples contain event mentions whose identity and position can be important to the underlying relationship expressed by the example. Applying perturbations without identifying these spans can unintentionally modify an important event, alter the semantics of the example, or interfere with the relationship that the example is intended to represent.

This model provides token-level event boundaries that can be used by the adversarial data-generation pipeline to locate event expressions and distinguish them from surrounding contextual text.

The detected spans provide structural information that can be used when constructing controlled adversarial variants of the original inputs.

The model therefore serves as an auxiliary component of the adversarial dataset generation pipeline. It is not intended to represent the primary model being evaluated for reasoning or generalization.

---

Role in the Adversarial Generation Pipeline

At a high level, the model is used as follows:

1. An original text example is provided to the model.
2. The model performs token-level event expression recognition.
3. Event spans are identified using the B-E and I-E labels.
4. The detected spans provide information about where event expressions occur in the input.
5. This information is used during construction of adversarial variants of the original examples.

The model therefore separates event-span detection from the subsequent adversarial transformation process, allowing the generation pipeline to operate with explicit information about the locations of events in the source text.

---

Usage

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

model_id = "mdg-nlp/event-ner-bert-base-cased"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForTokenClassification.from_pretrained(model_id)

classifier = pipeline(
    "token-classification",
    model=model,
    tokenizer=tokenizer
)

text = "The patient developed a fever before receiving treatment."

predictions = classifier(text)

print(predictions)

---

Limitations

• Event expressions that differ substantially from those encountered during fine-tuning may not be identified correctly.
• Tokenization can affect predicted event boundaries.
• Complex, overlapping, or ambiguous event mentions may result in incomplete or incorrect spans.
• The model is intended primarily as a supporting component for adversarial dataset construction.
• Predictions should not be interpreted as event-relation or temporal-relation predictions.

---

Research Context

This model was developed as part of research studying the behavior and generalization of NLP models under adversarial and other distributional changes.

Within that research pipeline, the model provides event-expression span information needed during the construction of adversarial examples. The resulting adversarial data can then be used independently to evaluate model behavior under systematically perturbed inputs.

---

Contact

Author: Arnav Verma
Affiliation: New York University (2026–Present), University of North Texas (2023–2026)

---

Citation

If you use this model in your research, please cite the associated work when available.