Semantic-role-labeling-progress

Semantic role labeling is a task of identify the predicate-argument structure in a sentence. It basically consists of 4-subtasks

• Identification of Predicate
  • Predicates can be of following types:
    • Verbial
    • Nomial
• Predicate Sense Disambiguation
• Argument identification
  • Arguments can be of two types:
    • Span based
    • Dependency based: where only syntactic head of the argument span is identified.
• Argument classification

Assumption

Most of the research works in SRL makes the following assumptions:

• Gold: Sentence boundary detection
• Gold: Sentence tokenized
• Gold: predicate location

In practical scenarios these assumptions does not hold and have a negative impact on the overall performance of a model.

Properties

Because of the complexity of the SRL task it is not always immediate to conclude that a particular model is State-of-the-Art. For example: Comparing a SRL model trained on x fine-tuned transformer model with a SRL model trained on y fine-tuned transformer model is not fair. Therefore, in this work we formalize the process of comparing different SRL models based on the following properties:

• Which dataset is used? CoNLL09, CoNLL05, CoNLL12, FrameNet
• Evaluation Metric: eval05, eval09
• compute predicate
• Computed predicate Sense: yes/NO
• Overall score: yes/no
• Separate score for predicate sense disambiguation?
• Separate score for argument classification?
• Whether syntax information is used in training SRL model?
• Type of Word embeddings used?
  • Fine-tunned embeddings
    • LMs
  • Static embeddings
    • LMs
    • ElMO
    • GLove
    • Seena
• Type of encoders
  • BiLSTMS
  • BiLSTMS + Attentions
  • NO encoder
• Model ensemble?

SRL Research

Syntax based

• [LREC 2022] Syntax-driven Approach for Semantic Role Labeling

Syntax agnostic

Multilingual

SRL datasets

Gold

• [CoNLL2009]
• [CoNLL2005]
• [CoNLL2012]

Silver

• [UP2.0]

SRL for downstream applications

Machine Translation

• [LREC 2022] Using Semantic Role Labeling to Improve Neural Machine Translation

Natural Language Inference

• [NAACL SUKI 2022] Is Semantic-aware BERT more Linguistically Aware? A Case Study on Natural Language Inference
• [Stanford CS224] Using Semantic Role Labeling to Combat Adversarial SNLI
• [ACL REPL4NLP 2020] Joint Training with Semantic Role Labeling for Better Generalization in Natural Language Inference
• [AAAI 2020] SemBERT: Semantics-aware BERT for Language Understanding

Question Answering

• [NAACL SemEval 2022] Hybrid Question Answering Using Semantic Roles

Content Moderation and verification

• [ACL FEVER 2022] A Semantics-Aware Approach to Automated Claim Verification
• [ACL Constraint 2022] Are you a hero or a villain? A semantic role labelling approach for detecting harmful memes.

Populating ontologies

• [Springer AI Law 2021] Populating legal ontologies using semantic role labeling
• [2018 ] Applying semantic role labeling and spreading activation techniques for semantic information retrieval

Other applications

• [Proposition Acquisition with SRL] https://github.com/pruizf/pasrl
• [Scientific Reports 2022] Computing semantic similarity of texts based on deep graph learning with ability to use semantic role label information
• [NAACL 2018] SRL4ORL: Improving Opinion Role Labeling using Multi-task Learning with Semantic Role Labeling
• [ Springer AI] Semantic role labeling for knowledge graph extraction from text

Sentiment Inference

• [KONVENS 2022] Semantic Role Labeling for Sentiment Inference: A Case Study