[Paper Review] Semantic Ambiguity and Perceived Ambiguity

Massimo Poesio|ArXiv.org|May 16, 1995

Logic, Reasoning, and Knowledge48 references76 citations

TL;DR

This paper develops a linguistically and cognitively grounded theory of semantic ambiguity and perceived ambiguity using underspecified representations, formalized via Default Logic to model defeasible reasoning in discourse interpretation. It resolves the combinatorial explosion problem in NLP by avoiding full enumeration of readings, instead relying on context-sensitive inference constrained by human-like disambiguation principles.

ABSTRACT

I explore some of the issues that arise when trying to establish a connection between the underspecification hypothesis pursued in the NLP literature and work on ambiguity in semantics and in the psychological literature. A theory of underspecification is developed `from the first principles', i.e., starting from a definition of what it means for a sentence to be semantically ambiguous and from what we know about the way humans deal with ambiguity. An underspecified language is specified as the translation language of a grammar covering sentences that display three classes of semantic ambiguity: lexical ambiguity, scopal ambiguity, and referential ambiguity. The expressions of this language denote sets of senses. A formalization of defeasible reasoning with underspecified representations is presented, based on Default Logic. Some issues to be confronted by such a formalization are discussed.

Motivation & Objective

• To establish a connection between underspecification in NLP, semantic ambiguity, and human ambiguity processing.
• To address the combinatorial explosion puzzle—where syntactic and scopal ambiguities generate exponentially many readings—by avoiding full enumeration of interpretations.
• To develop a theory of underspecification from first principles, grounded in model-theoretic semantics and cognitive plausibility.
• To formalize perceived ambiguity as the result of conflicting defeasible inferences rather than inherent semantic multiplicity.
• To unify scope disambiguation, reference resolution, and discourse interpretation under a single inferential framework.

Proposed method

• Defines semantic ambiguity model-theoretically as the existence of multiple senses for a linguistic expression.
• Introduces an underspecified language whose expressions denote sets of possible senses, avoiding commitment to a single interpretation.
• Applies Default Logic to formalize defeasible reasoning with underspecified representations, enabling context-sensitive interpretation.
• Imposes constraints on inference rules via the Anti-Random Hypothesis and the Condition on Discourse Interpretation to ensure cognitive plausibility.
• Derives the Condition on Discourse Interpretation from Pinkal’s precisification imperative, linking it to the resolution of H-type (referential) ambiguity.
• Extends the framework to handle lexical, scopal, and referential ambiguity, with preliminary treatment of syntactic ambiguity.

Experimental results

Research questions

• RQ1 How can a theory of underspecification be developed that is both linguistically accurate and cognitively plausible?
• RQ2 What distinguishes semantic ambiguity (multiple senses) from perceived ambiguity (conflicting inferences) in discourse interpretation?
• RQ3 How can defeasible reasoning with underspecified representations avoid the combinatorial explosion problem in NLP systems?
• RQ4 What constraints are necessary to ensure that interpretation processes are non-random and context-sensitive?
• RQ5 How do scope disambiguation and reference resolution interact in a unified interpretation framework?

Key findings

• The theory successfully models perceived ambiguity as arising from conflicting defeasible inferences, not from inherent semantic multiplicity.
• The use of underspecified representations avoids the need to generate all possible readings, thus mitigating the combinatorial explosion problem.
• The Condition on Discourse Interpretation, derived from the precisification imperative, ensures that H-type ambiguity is resolved through context-sensitive inference.
• The formalization using Default Logic provides a principled mechanism for defeasible reasoning with underspecified representations.
• The framework supports a unified treatment of lexical, scopal, and referential ambiguity, with preliminary extensions to syntactic ambiguity.
• The theory underpins the SAD-93 system used in the TRAINS-93 demo, demonstrating practical applicability in task-oriented dialogue systems.