Language models trained on tokenized text still reliably produce morphemes whose form depends on phonology, but it was unclear whether this relies on memorization or rule-like generalization. The study shows that for the English indefinite article a/an, the phonological condition is encoded along a single linear direction in trigger-token embeddings, causally drives article selection in token-level wug tests, and is used by the model to forecast the upcoming trigger token’s phonological feature for article choice. The authors further investigate whether this rule-like generalization extends to allomorph selection in other languages and to explicit phonological judgment, offering a mechanistic account that separates generation-time ability from metalinguistic judgments.
arXiv:2608. 04021v1 Announce Type: cross Abstract: Cloze-style probes that vary how often a target token appears implicitly assume that more copies of a target affect prediction the same way regardless of where the readout slot sits.
The paper investigates how language models avoid overgeneralizations by distinguishing between two types of indirect negative evidence: preemption and entrenchment. Through controlled rearing experiments on models trained on child‑caregiver conversations, the authors find that models do not exhibit verb‑specific preemption but show weak abstract preemption. Analysis of training dynamics suggests that competing structures act as indirect positive evidence rather than negative in the verb‑specific condition.
The paper studies structural priming in language model production by conducting controlled sentence‑completion experiments on dative constructions. Results show that language models exhibit priming effects, especially when sentences are semantically coherent, with stronger relative increases for double‑object datives and larger absolute increases for prepositional‑object datives. The study also finds that primed completions involve more lexico‑semantic repetition, indicating that priming operates across syntactic, lexical, and semantic levels.
arXiv:2608.29921v1 Announce Type: cross Abstract: The output of a Language Model can be tampered with \emph{while} the model is writing it. A simple test can thus be constructed by evaluating the mod...
arXiv:2606. 00230v1 Announce Type: new Abstract: Grokking, the phenomenon in which neural networks generalize long after fitting their training data, has been studied in supervised settings on many epochs.
arXiv:2606. 07555v1 Announce Type: cross Abstract: Glossaries, technical specifications, and system prompts routinely ask language models to use familiar words in unfamiliar ways.
arXiv:2608. 00523v2 Announce Type: replace-cross Abstract: The linguistic notion of state has traditionally been restricted to the construct (annexation) state of Afroasiatic languages and treated as a language-specific morphosyntactic phenomenon.
arXiv:2607. 13568v1 Announce Type: cross Abstract: Can a language model estimate its familiarity with an entity before generating an answer?
arXiv:2603. 18482v2 Announce Type: replace-cross Abstract: Standard decoding strategies for text generation, including top-$k$, nucleus sampling, and contrastive search, select tokens based on likelihood, restricting outputs to high-probability regions.
arXiv:2605. 23080v2 Announce Type: replace Abstract: Feature attribution methods promise to identify which input features matter for a model output.
The study investigates why language models exhibit systematic performance gaps across English dialects, a phenomenon termed the "dialect tax." Using parallel dialect corpora that preserve meaning while altering surface form, the authors confirm that models treat Standard American English and dialectal texts as semantically equivalent, yet find representational disparities that persist through tokenization, pre‑training, post‑training, and inference. Even a character‑level tokenizer does not eliminate input/output asymmetries or accuracy gaps, and dialect pairs produce more divergent gradient updates than unrelated Standard texts, indicating that dialectal content is harder for models to learn.
The paper argues that language operates with two parameters: amplitude, which measures how often words co‑occur, and phase, a signed relational factor that determines how co‑activated meanings combine and can reverse a meaning’s contribution. Unlike amplitude, phase is not captured by standard word embeddings or transformer attention weights and is indexed to individuals and dyadic interactions. The authors propose six empirical predictions to test phase’s role and suggest that future language models should incorporate agent‑indexed, phase‑bearing semantic states.