Evidence-Grounded QA System

An Evidence-Grounded QA System is a question answering system that is an explainable NLP system implementing evidence-aware QA algorithms to solve evidence-grounded question answering tasks.

• AKA: Evidence-Based Question Answerer, Interpretable QA System.
• Context:
  • It can typically implement Evidence Retrieval Modules before answer extraction.
  • It can typically maintain Evidence-Answer Alignments through attention mechanisms.
  • It can typically provide Provenance Tracking for answer sources.
  • It can typically generate Confidence Scores based on evidence quality.
  • It can typically support Answer Explanations via evidence highlighting.
  • ...
  • It can often employ Dense Passage Retrievers for evidence selection.
  • It can often utilize Cross-Attention Mechanisms between question and evidence.
  • It can often incorporate Reranking Modules for evidence optimization.
  • It can often implement Multi-Granularity Processing from sentence to document level.
  • ...
  • It can range from being a Pipeline QA System to being an End-to-End QA System, depending on its architectural design.
  • It can range from being a Retrieval-Based QA System to being a Generation-Based QA System, depending on its answer production method.
  • ...
  • It can process Natural Language Questions with document collections.
  • It can output Answer Texts with evidence passages.
  • It can integrate with Search Engines for document retrieval.
  • It can be evaluated by Evidence-Grounded QA Performance Measures.
  • ...
• Example(s):
  • BERT-Based QA Systems with passage ranking, such as:
    • DPR-Reader System combining retrieval and reading.
    • REALM System with knowledge retrieval.
  • Multi-Stage QA Systems, such as:
    • DrQA System with document retrieval and reading comprehension.
    • ORQA System with open retrieval.
  • Interactive QA Systems, such as:
    • ConversationalQA System maintaining evidence context.
    • Clarification QA System requesting additional evidence.
  • ...
• Counter-Example(s):
  • Generative QA Systems without evidence grounding.
  • Template-Based QA Systems using fixed patterns.
  • Knowledge Base QA Systems querying structured data.
• See: Explainable QA System, Reading Comprehension System, Evidence Retrieval System, Interpretable NLP System.

---