Semantic Role Labeling: Who Did What to Whom

Module: Natural Language Processing | Difficulty: Advanced

PropBank Args

SRL as Sequence Labeling

Evaluation

import torch
import torch.nn as nn

class SRLModel(nn.Module):
    def __init__(self, bert_model, n_roles):
        super().__init__()
        self.bert = bert_model
        self.role_classifier = nn.Linear(768, n_roles)
    def forward(self, input_ids, attention_mask):
        outputs = self.bert(input_ids, attention_mask=attention_mask)
        hidden = outputs.last_hidden_state
        return self.role_classifier(hidden)

Research Insight: SRL provides shallow semantic understanding of text. Neural models improved SRL by 10+ F1 points by learning contextual representations. The key insight is that SRL benefits from pre-trained language models because argument structure is partially captured by language modeling.