arXiv

Noisy memory encoding explains negative polarity illusions

Title: How Noisy Memory Encoding Drives Negative Polarity Illusions

Abstract

Sentences such as "The authors that no critics recommended have ever received acknowledgment for a best-selling novel" are frequently judged as grammatically acceptable, despite being technically incorrect. The error stems from the fact that the negative polarity item "ever" is not properly licensed in its current context. This phenomenon, known as a "negative polarity illusion," is explained in this study by the lossy context surprisal theory proposed by Hahn et al. (2022), which posits that individuals possess an imperfect encoding mechanism for complex sentences.

We hypothesize that memory representations of the determiners within both the main-clause and embedded-clause subjects are often weak. Consequently, readers may mentally substitute one determiner for another, a swap that could render the use of "ever" grammatically valid. Furthermore, we suggest that the strength of this illusion correlates with the semantic similarity of the determiners involved in such exchanges.

To test this, we conducted acceptability judgment tasks using six novel determiner pairs, such as "few" versus "many" and "few" versus "most." The results confirmed our hypothesis: a sentence like "Many authors that few critics recommended have ever received acknowledgment for a best-selling novel" elicited a significantly stronger illusion than the standard example, even when participants were not under time pressure. These findings reinforce the view that human language processing is both resource-rational and imperfect. Facing the constraints of working memory, individuals rationally reconstruct the most probable meaning from noisy linguistic input to streamline subsequent processing.


Source: arXiv Generated at: 2026-06-04 00:00:00 UTC

Related Articles

AI Concentration Risk Is the Problem: 3-Minutes MLIV
Bloomberg

AI Concentration Risk Is the Problem: 3-Minutes MLIV

The article argues that AI concentration risk, rather than the technology itself, is the primary concern. It highlights ...

Reuters

Foxconn announces strategic collaboration with Intel on next-gen AI infrastructure

Foxconn and Intel announced a strategic partnership to develop next-generation AI infrastructure. This collaboration aim...

SpaceX Seeks to Raise $75 Billion in Record IPO (Video)
Bloomberg

SpaceX Seeks to Raise $75 Billion in Record IPO (Video)

SpaceX aims for a record $75 billion valuation through an initial public offering. This historic IPO marks a significant...

Broadcom AI Chip Outlook Disappoints Investors
Bloomberg

Broadcom AI Chip Outlook Disappoints Investors

Broadcom’s AI chip projections disappointed investors, dampening market sentiment. The outlook fell short of expectation...

Reuters

Europe's tech 'liberation day'? Computer says not yet

Europe’s expected tech breakthrough remains unrealized, as current systems indicate that a true "liberation day" has not...

Hiranandani Group CEO on Powering India's Digital Future
Bloomberg

Hiranandani Group CEO on Powering India's Digital Future

Hiranandani Group CEO discusses driving India's digital transformation.