SciBorg

Automating Discourse Analysis in Scientific Chemistry Publications

The SciBorg UK initiative, spearheaded at the University of Cambridge, established foundational natural language processing architectures to extract rhetorical structures, chemical entities, and experimental findings directly from peer-reviewed scientific papers. By deploying automated discourse analysis across 15 semantic zones, the system bridges raw chemistry text and structured relational knowledge bases.

Architecture Comparison: Pipeline Stages and Parsing Approaches

End-to-End Extraction Workflow for Chemistry Corpora

1. Step 1: Document normalization converting raw PDF and XML feeds into standardized SciXML corpora.
2. Step 2: Lexical analysis applying OSCAR4 for automated identification of chemical IUPAC names and ontology concepts.
3. Step 3: Rhetorical zone classification assigning Core Scientific Concepts (CoreSC) across each sentence.
4. Step 4: Relational knowledge graph synthesis exportable for automated citation synthesis and meta-analysis.

Domain-Specific Challenges in Chemical Natural Language Processing

Ambiguities arise frequently within abbreviated IUPAC strings, polymer compositions, and inline experimental protocols. Standard syntactic parsers degrade when processing nested parenthetical compounds exceeding 4 embedded tiers or non-standard typographic symbols in synthesis workflows.