referencebuilding

Referencebuilding is the process of creating and identifying references to entities within a discourse or data space. It involves selecting referring forms—such as pronouns, definite descriptions, demonstratives, or proper names—and linking them to intended referents in a way that is coherent and efficient for the audience. In this sense, referencebuilding connects linguistic expression with referential intention.

In human communication, referencebuilding relies on context, shared knowledge, discourse history, and social norms to determine which entity is being referred to and which form best suits the interaction. Deixis anchors expressions to the immediate situation, while anaphora links later mentions to earlier referents. The choice of form reflects factors such as salience, formality, and communicative goals, contributing to cohesion and reducing ambiguity.

In natural language processing, referencebuilding encompasses coreference resolution and referring expression generation. Coreference resolution identifies all mentions that refer to the same underlying entity within a text, whereas referring expression generation designs expressions that uniquely identify a target given the surrounding context and the listener’s knowledge. These tasks enable machines to understand and produce coherent, context-appropriate language.

In data management and knowledge integration, referencebuilding involves linking entities across sources, resolving duplicates, and assigning stable identifiers. Techniques include entity disambiguation, record linkage, and the use of global identifiers or schema mappings to create consistent representations of the same item across systems.

Applications of referencebuilding span dialog systems, text understanding, search, and knowledge graphs. Challenges include linguistic ambiguity, polysemy, multilingual variation, and changing contexts. Ongoing work aims to model context, pragmatics, and user-specific conventions more robustly to improve referential clarity and interoperability.