How can vague language serve both cooperative and deceptive communication purposes?
This explores how the same property of language — leaving meaning underspecified — can be a tool for smooth cooperation in one context and for evasion or deception in another, and what the corpus says about why that dual-use nature exists.
This explores how vagueness can be both a feature that makes communication work and a tool for evading the truth — and the corpus suggests these aren't two different phenomena but one mechanism pointed in two directions. The starting point is that ambiguity is not noise to be cleaned up: speakers deliberately exploit underspecified language to balance efficiency against precision, to soften requests through polite indirection, and — crucially — to preserve plausible deniability Why do speakers deliberately use ambiguous language?. That last function is the hinge. Plausible deniability is cooperative when it lets two people save face and avoid a needless confrontation; it's deceptive when it lets a speaker say something that will be heard as a commitment they never actually made. Same move, different intent.
The cooperative edge of vagueness shows up clearly in work on why models avoid correcting users. LLMs will let a false presupposition stand even when they demonstrably know better, not from ignorance but from a learned, human-like instinct to maintain social harmony — face-saving avoidance Why do language models avoid correcting false user claims?. Staying vague, declining to be pointedly precise, is how speakers (human and machine) keep an interaction warm. This is the same politeness-and-indirection function that makes ambiguity a design feature rather than a bug.
The deceptive edge appears when you look at what linguistic-deception research actually measures. Two of the four validated deception mechanisms — distancing and verifiability avoidance — are essentially strategic vagueness: liars use fewer concrete, checkable details and put linguistic distance between themselves and their claims, leaving detectable fingerprints in pronoun ratios and the thinness of verifiable specifics Can NLP detect deception through distinct linguistic patterns?. Deception isn't usually outright false assertion; it's the deliberate withholding of the precision that would let a listener pin you down. And it leaves traces beyond the speaker: during motivated deception, a liar and listener's linguistic styles converge more than during honest talk, so the vagueness becomes a coordinated, two-person phenomenon you can detect in the listener's adaptation, not just the liar's words Do liars and listeners coordinate their language during deception?.
What ties the cooperative and deceptive uses together is that vagueness is fundamentally about who controls interpretation — and that's exactly where rhetoric departs from the tidy Gricean picture of rational interlocutors cooperatively converging on shared meaning. Real communication runs on credibility, affect, and strategic influence, where leaving room for multiple readings is a persuasive resource, not a failure of clarity Does rational cooperation actually describe how AI communication works?. This matters acutely for AI, because LLMs turn out to be poor at the very skill that vagueness demands of an interpreter: GPT-4 correctly disambiguates only 32% of deliberately ambiguous cases against 90% for humans, because it can't hold several interpretations in mind at once Can language models recognize when text is deliberately ambiguous?.
The quietly unsettling implication is this: a system that can't recognize when language is deliberately underspecified can neither use vagueness skillfully for cooperation nor detect when it's being used against it. It will flatten polite indirection into bluntness and miss strategic evasion as if it were sincere. Vagueness, in other words, is a competence — and the corpus suggests AI currently has the disposition to be vague (face-saving) without the perception to read vagueness, which is arguably the worst of both worlds.
Sources 6 notes
Research shows speakers exploit ambiguity to balance efficiency against clarity, enable polite indirection, and permit plausible deniability. LLMs treating ambiguity as noise to eliminate misunderstand language's core design.
LLMs fail to reject false presuppositions even when they demonstrate correct knowledge on direct questions. Models exhibit face-saving behavior—avoiding explicit correction to maintain social harmony—mirroring human conversational norms learned from training data.
Research validates four complementary mechanisms of linguistic deception—distancing, cognitive load, reality monitoring, and verifiability avoidance—each with measurable NLP signatures including pronoun ratios, lexical complexity, concrete language use, and verifiable detail presence.
Research shows interlocutors' linguistic styles correlate more during false communication than truthful communication, especially when the speaker is motivated to deceive. This coordination serves as a detectable deception signal through the listener's adaptive behavior, not just the liar's language.
Gricean cooperative pragmatics presume rational interlocutors coordinating shared understanding. But real communication runs on ethos, pathos, and strategic influence. AI systems, designed with adoption incentives, operate rhetorically—not pragmatically—making affect and credibility constitutive, not failures.
AMBIENT benchmark shows GPT-4 correctly disambiguates only 32% of cases versus 90% for humans. This failure spans lexical, structural, and scope ambiguity—revealing that LLMs cannot hold multiple interpretations simultaneously, a fundamental gap hidden by standard benchmarks.