Abstract
This paper introduces a formal computational framework for "Argumentative Semantics," which aims to extract the underlying persuasive structure of natural language. By formalizing how linguistic markers—such as "but," "unfortunately," and "little"—constrain a sentence's meaning, the author provides a method to calculate speaker commitment and argumentative orientation. The framework integrates polyphony theory with topoi-based warrants to model how language steers listeners toward specific conclusions.
Introduction
Standard natural language processing often prioritizes literal or "informative" content. However, human communication is inherently argumentative; we select specific words to influence a listener's perspective. For instance, the semantic difference between "I have little money" and "I have a little money" is not found in the factual amount, but in the argumentative direction: the former suggests poverty (negative), while the latter suggests resourcefulness (positive).
Traditionally, these nuances are relegated to pragmatics—context-dependent usage. This paper argues that these constraints are intrinsic to linguistic structure and proposes a formal methodology to compute them prior to contextual analysis.
Methods
The proposed framework builds upon Oswald Ducrot's "Integrated Pragmatics" and utilizes three core components to model argumentation:
- Polyphony: This theory decomposes an utterance into distinct "voices." It distinguishes the talking subject (the physical speaker) from the speaker (the persona) and the utterers (entities responsible for specific semantic content). The model tracks the speaker's commitment to these contents using three values: Identification (ID), Opposition (OPP), and Partial Commitment (PC).
- Topoi: These are non-logical warrants or commonplaces that link concepts (e.g., linking "wealth" to "purchasing power"). Words encapsulate these topoi, which possess both direction (positive/negative) and strength.
- A-Structures and Φ-Structures:
- The A-structure is a syntactic tree identifying operators (e.g., "unfortunately"), connectives ("but"), and modifiers ("a little") linked to terminal sub-sentences (TSS).
- The Φ-structure is a mathematical function that transforms "argumentative cells" (topos, direction, strength, commitment) of sub-sentences based on the applied operators.
The computation follows a bottom-up process, starting with the literal potential of sub-sentences and applying functional transformations to arrive at the final "signification" of the complete utterance.
Key Insights and Ideas
- Linguistic Argumentation: Argumentative power is a constraint imposed by a sentence's linguistic superstructure, not merely an inference drawn from context.
- The "But" Shift: The connective "but" is formalized as a tool for generating alternatives. In a "P1 but P2" structure, the speaker typically assigns "Partial Commitment" to P1 and "Identification" to P2, ensuring the overall argument aligns with P2 while acknowledging P1.
- Modifiers as Directional Toggles: Operators like "little" versus "a little" act as toggles for "topical direction." "Little" can invert a positive orientation into a negative one (anti-orientation).
- Requirement for "Odd" Contexts: The framework explains semantic dissonance. If a linguistic structure forces an interpretation contradicting common sense (e.g., framing a robbery as "lucky"), the model identifies this as a requirement for a specific, unusual context to validate the sentence.
Results
The paper offers a qualitative proof-of-concept for a semantic module capable of:
- Formalizing the "signification" of complex sentences by treating words as functional operators.
- Computationally tracking speaker commitment through polyphonic values.
- Supporting "top-down" processing, where global sentence effects (like sarcasm) impose constraints on the interpretation of individual sub-sentences.
Limitations and Future Work
Despite its formal strengths, the model currently relies on manual lexical descriptions for the argumentative properties of words. Creating a truly compositional signification for sub-sentences remains a challenge. Additionally, the lack of an exhaustive database of "topoi" limits broad application. Future research must also better integrate these argumentative structures with "informative" (fact-based) and "illocutionary" (speech act) semantic elements.
Summary updated