Safety-incident reports are written with deliberate epistemic caution: A reporter marks what was seen, what was inferred, and what remains unsure, and a dropped hedge can turn a tentative observation into an apparent cause. As large language models take over summarizing such reports for triage, indexing, and search, whether they preserve that caution becomes a property of the safety record itself. This study asks whether models keep a reporter's expressed uncertainty when they summarize, testing two hypotheses fixed before any summary was generated: that models systematically harden uncertainty by default and that an explicit instruction to preserve it overcorrects into softening. Six open-weight models summarized 42 real incident reports, 30 from the United States Aviation Safety Reporting System and 12 from the United Kingdom Marine Accident Investigation Branch, under a plain prompt, a preserve prompt, and a paraphrase of it. Uncertainty change was measured two ways: at the passage level, by a deterministic epistemic-marker lexicon, and at the claim level, by aligning those markers between source and summary, each validated against manual annotation and cross-checked against two independent language-model judges. Under plain summarization, all six models hardened the reporter's uncertainty (claim-level hardening 0.59–0.85), dropping explicit hedges almost entirely (retention about 0.08) while epistemic modal verbs survived far better. The preserve instruction reversed the pattern into over-softening: In a cluster-robust logistic regression, it carried an odds ratio of 0.12 (95% CI [0.03, 0.49], p = 0.003) and held once summary length was controlled, so the change reflects expressed certainty rather than compression. In practical terms, the instruction lowered the predicted probability of hardening from about 0.43 to 0.13 (Cohen's h = 0.69), and the effect held in mixed-effects models and under adverse imputation of the missing outputs. The pattern was directionally consistent across the tested models and both instruction wordings. Automated safety triage that relies on model summaries should measure epistemic fidelity directly and not treat a single preserve instruction as a solution.
Veronica Mangiaterra, Chiara Barattieri di San Pietro, Paolo Canal, Valentina Bambini
Metaphors are a distinctive feature of literary language, yet they remain less studied experimentally than everyday metaphors. Moreover, previous psycholinguistic and computational approaches overlooked the temporal dimension, although many literary metaphors were coined centuries apart from contemporary readers. This study innovatively applies tools from diachronic distributional semantics to assess whether the processing costs of literary metaphors varied over time and genre. Specifically, we used the Word2Vec algorithm and period-specific literary and nonliterary corpora, for a total of 124 million tokens, to generate word embeddings. Using these embeddings, we modeled changes in the semantic similarity between topics and vehicles of 515 19th-century literary metaphors, taking this measure as a proxy for metaphor processing demands. Overall, semantic similarity, and hence metaphor processing demands, remained stable over time. However, genre played a key role: metaphors appeared more difficult (i.e., lower topic-vehicle similarity) in modern literary contexts than in 19th-century literature, but easier (i.e., higher topic-vehicle similarity) in today's nonliterary language (e.g., the web) than in 19th-century nonliterary texts. This pattern was further shaped by semantic features of metaphors' individual terms, such as vector coherence and semantic neighborhood density. Collectively, these findings align with broader linguistic changes in Italian, such as the stylistic simplification of modern literature, which may have increased metaphor processing demands, and the high creativity of the web's language, which seems to render metaphor more accessible.
The part-whole relations of body-part lexemes, or “body partonymy,” might seem like a key source of evidence for shared conceptual systems in support of embodiment theories of cognition, but widespread variation in this lexical field across world languages calls this assumption into question. Recent studies focus on radical mismatches and random overlap relative to isolated frames of contrast such as body joints, reference languages, and ad hoc hierarchies. This study takes a fresh approach to comparative research on body partonymy by focusing on analogous relations between upper- and lower-body membership sets or “transverse analogous paradigms” (TAP), hypothesizing that most world languages feature strong evidence of such relations. Focusing on upper and lower “limb” (arm/leg) relationships, I test this hypothesis on an original database of 50 world languages built from detailed lexicographic sources quota sampled for broad distribution in terms of physical geography, population size, and genetic linguistic diversity. The analysis operationalizes four degrees of evidence: transverse parallel conceptions, with or without arm/leg matching, followed by transverse colexifications, and transverse paradigm sets. Coding for these four degrees of evidence reveals moderate evidence for TAP in all languages sampled, with 76% (n = 38) showing comprehensive indices (four out of four degrees), an additional eleven (22%, 98% total) showing strong indices (three out of four degrees) and only one language showing modest evidence (two out of four degrees). Findings demonstrate the value of reorienting comparative body partonym studies to the intrinsic organization of upright posture itself, with potentially broader implications for theorizing the cognitive foundations of grammar and the origins of advanced conceptual blending.
Reading fluency is central to reading development, yet its conceptual boundaries and changing relations with word recognition and comprehension remain unsettled. This conceptual analysis proposes that fluency development changes how processing time is organized across connected text. As word identification consolidates, timing is hypothesized to become increasingly responsive to linguistic integration demands and reading goals. Fluent performance can therefore include selective local slowing followed by recovery of continuity. The framework draws on established accounts of automaticity, synchronization, word-to-text integration, and strategic regulation. Fluency is characterized through accuracy, temporal continuity and pace regulation, and, in oral reading, prosodic organization. Connected-text coordination is proposed to explain the temporal relationships among word identification, linguistic grouping, and local meaning updating. Three overlapping configurations organize the relative prominence of lexical constraints, integration demands, and regulation by meaning and reading goals. Their number and developmental ordering remain hypotheses. Reading experience can support lexical consolidation and better calibrated linguistic expectations, with oral-language resources, morphological development, compensatory profiles, and orthographic variation constraining the account. The primary hypothesis predicts that, as word identification consolidates under comparable reading conditions, the association between global reading rate and comprehension will weaken, while the association between demand-sensitive local timing and comprehension will remain stable or strengthen. This joint prediction follows from the proposed change in processing constraints, beyond faster word identification alone. Secondary hypotheses address goal-dependent rereading and generalization across oral and silent reading. Evaluation requires independently specified demands and outcomes, longitudinal measurement, and comparison with existing accounts. The framework's proposed contribution is a developmental explanation linking the organization of processing time to the changing informativeness of fluency indicators. Its explanatory value depends on these commitments distinguishing it from alternative accounts.
There are two types of click sounds that have been studied in linguistics. The first are those that form part of the consonant inventory of certain African languages, such as Zulu and Xhosa. In these languages, clicks are used phonemically. The second group of click sounds exist in many - if not all - languages, but they are used paralinguistically, that is, they convey pragmatic and discursive functions. For example, a click can convey exasperation, surprise, or the idea that the speaker is searching for what to say. Although these latter clicks are common, it is not clear to what extent speakers are aware of them. Vigil and Pinto (2020) found that English speakers are hardly able to detect paralinguistic clicks in stretches of semi-spontaneous speech. The present study uses the same paradigm and stimuli in an online setting, and tests both native English speakers and multilingual speakers of English and one or more click languages. The results show that people who grew up speaking a click-based language are better than English speakers at spotting some of these paralinguistic clicks.
