Stablereferring

Stablereferring is a term used in linguistics and natural language processing to describe the set of strategies and phenomena that preserve a consistent referent for an entity across multiple sentences or discourse segments. The goal is to maintain identifiability and reduce ambiguity as a text unfolds. Stable referring relies on a combination of linguistic choices, such as the timing of reintroducing nouns, the use of pronouns with clear antecedents, and the maintenance of a stable coreference chain across sentences.

In humans, stability emerges from discourse planning and coherence pressure, with readers and listeners tracking entities across the narrative. In computational systems, stablereferring is implemented through coreference resolution, entity tracking, and generation strategies that preserve a consistent internal representation of entities over the course of a document. Common challenges include introducing new referents mid-discourse, multiple potential antecedents, and ellipsis or metaphor that can blur identity.

Evaluation typically combines automatic coreference metrics with discourse-level stability assessments, evaluating how often the same referent is maintained and how often pronouns are correctly linked to their antecedents across longer passages. Applications include long-form text generation, legal or technical drafting, summarization, and conversational agents that must sustain coherent references across turns.

Example: "The engineer reviewed the data. She noted irregularities." Stable referring is achieved when "she" unambiguously refers to the engineer, preserving identifiability across the sentences. Related topics include coreference resolution, discourse coherence, and entity tracking.