R. Edirisinghe

03 Research architecture

A connected system for recovering relation beyond surface similarity

With the transformation problem visible, the research moves from texts and representations to candidates, ranked relations, evidence, and visual exploration.

Doctoral research object pipelineTexts become representations, a candidate space, ranked relations, evidence, and visual exploration through the operations represent, retrieve, judge, and explore.TEXTSREPRESENTATIONSCANDIDATESPACERANKEDRELATIONSEVIDENCEVISUALEXPLORATION
Object 01 / TEXTS

Source and target passages enter the pipeline as texts situated in context.

Within REPRESENT: Computational representations for relations that persist across paraphrase, adaptation, obfuscation, allusion, historical variation, and related transformations.
Layered textual-relation research pipelineResearch objects move upward through texts, embedding representations, candidate retrieval, ranked relations, evidence, and visual exploration.INPUTTEXTSREPRESENTREPRESENTATIONSRETRIEVECANDIDATESPACERANK / JUDGERANKEDRELATIONSEVIDENCEEVIDENCEEXPLOREVISUALEXPLORATION
Layer 01 / TEXTS

Source and target passages enter the pipeline as texts situated in context.

  1. TEXTS

    Source and target passages enter the pipeline as texts situated in context.

  2. REPRESENT REPRESENTATIONS

    Texts become computational representations that preserve signals relevant to relation.

  3. RETRIEVE CANDIDATE SPACE

    Retrieval narrows a large collection to passages that may warrant comparison.

  4. JUDGE RANKED RELATIONS

    Candidate relations are ordered for closer assessment rather than treated as equivalent.

  5. JUDGE EVIDENCE

    Judgement returns to the textual evidence that supports or complicates a relation.

  6. EXPLORE VISUAL EXPLORATION

    The resulting evidence can be explored across a corpus and returned to close reading.

NextCurrent activity