Title: AI Chatbots Generate Semantically Rich Yet Contentious Text

Original: arXiv:2606.02973v1 Announce Type: new

Abstract: Do the responses generated by artificial intelligence chatbots carry genuine meaning? Consider a standard interaction: a user queries Anthropic’s model, Claude, by asking, "What is the capital of Spain?" If Claude replies, "Madrid is the capital of Spain," does that statement possess its conventional semantic value, and is it a true proposition? For the vast majority of laypeople and AI developers, the answer is an obvious "yes." Conversely, a significant contingent of philosophers of language, linguists, and cognitive scientists contends that prevailing intentionalist theories of meaning yield the opposite result.

In an effort to align theoretical frameworks with the common-sense intuition that Large Language Model (LLM) outputs are meaningful, some theorists have proposed a radical "de-anthropomorphization" of language. This approach seeks to revise our conceptualizations of mental states, intentions, and semantic content. We propose an alternative perspective. While we agree that LLM outputs are meaningful, we argue that existing theories of human language are already sufficient to account for current chatbot behavior, requiring no modification.

We posit that the threshold for meaning is relatively low. Asserting that LLM outputs are meaningful does not necessitate attributing mental states, intentions, rationality, or the cognitive faculties required for human communication to these systems, nor does it demand any other anthropomorphic assumptions. Even among humans, who typically possess successful communicative intentions, language production frequently diverges from the speaker’s internal thoughts.

This perspective carries significant implications for how we theorize about and critically engage with both human linguistic expressions and synthetically generated text. Crucially, acknowledging that chatbots produce meaningful text does not imply an endorsement of their specific outputs, nor does it dictate whether the technology is beneficial, powerful, appropriate, or useful.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC