{"id":"31711752-859f-4c5c-bcbb-89f272003100","arxiv_id":"2605.09187","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Semantic role understanding partially emerges during language model pre-training, with linear probes on frozen representations achieving substantial performance that improves with scale but does not match fine-tuned models, and representations shifting toward more distributed forms at larger scales.","lead":"The paper finds that semantic role information like 'who did what to whom' is partially present in the frozen representations of decoder-only language models after pre-training alone, with linear probes showing substantial but incomplete performance that improves with scale yet remains below fine-tuned levels. This matters because it clarifies how much linguistic structure arises from unsupervised language modeling objectives versus task-specific adaptation.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.3","headline":"Linear probe accuracy does not directly establish a shift to more distributed internal representations with scale","rationale":"The reader's weakest assumption correctly flags probe interpretation but understates the specific gap for the 'distributed' clause of the strongest claim. The emergence-from-LM part is on firmer ground via scale trends and frozen-vs-fine-tuned comparison; the implementation-shift part requires an additional observable that the abstract and reader summary do not confirm is present.","tokens_in":1625,"tokens_out":346,"duration_ms":73898,"concrete_test":"In the methods/results sections, locate any quantification of distributedness (participation ratio, effective rank of role-related subspace, or linear-vs-MLP probe gap). If absent, train a small nonlinear probe (1-hidden-layer MLP) on the same frozen representations at each scale and compare accuracy delta to the linear probe; if the nonlinear advantage grows with scale, the distributed-shift claim is supported; if not, it weakens.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim asserts both emergence of semantic role structure from LM pre-training and a scale-dependent shift toward more distributed implementations. The reported evidence is linear probe performance on frozen decoder-only representations, which improves with scale yet remains below fine-tuned levels. This pattern supports partial linear decodability but does not isolate whether the underlying encoding becomes more distributed (e.g., higher effective dimensionality, lower sparsity, or reliance on nonlinear combinations) versus simply stronger or more redundant. Without an explicit metric for distributedness or a comparison (linear vs. nonlinear probes, dimensionality reduction, or unit ablation), the 'shift' component rests on an unstated inference from probe results alone.","agreement_with_reader":"partial"},"referee_report":{"model":"grok-4.3","summary":"The paper examines whether semantic role understanding emerges in decoder-only language models solely from pre-training. It freezes model representations across scales and trains linear probes to decode semantic roles (who-did-what-to-whom), comparing probe accuracy against fine-tuned baselines. Results indicate substantial role information is linearly decodable from frozen pre-trained representations, with accuracy improving as scale increases yet remaining below fully fine-tuned performance; the authors conclude that semantic role structure emerges from language modeling objectives but shifts toward more distributed internal representations with larger models.","tokens_in":1756,"tokens_out":473,"duration_ms":47342,"significance":"If substantiated with appropriate controls and metrics, the work would provide useful empirical evidence on the limits of unsupervised emergence of core linguistic structure in LMs, particularly the partial nature of semantic role encoding and its scale dependence. The probe-based methodology on frozen decoder-only models is a standard tool for such questions and could inform debates on what pre-training actually captures versus what requires task-specific adaptation.","major_comments":[{"comment":"Abstract and §3 (results on scale dependence): The claim that 'its internal implementation shifts toward more distributed representations as model scale increases' is not directly supported by the reported linear probe accuracies alone. Improved linear decodability with scale could reflect stronger, more redundant, or higher-dimensional encodings without necessarily indicating a change in distributedness; no explicit metric (e.g., effective dimensionality, sparsity measures, unit ablation, or linear-vs-nonlinear probe comparisons) is described to isolate this shift.","section":"Abstract / §3"},{"comment":"Abstract and methods: The central inference that linear probe performance on frozen representations indicates semantic role information 'already encoded' during pre-training lacks reported controls for probe capacity, random baselines, or statistical significance testing. Without these, it remains possible that probes introduce or amplify structure rather than purely extract pre-existing encodings, weakening the emergence claim.","section":"Abstract / Methods"}],"minor_comments":[{"comment":"The abstract provides no numerical performance values, exact model scales tested, or dataset details for the semantic role probing task, making it difficult to assess the magnitude of the reported improvements.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their constructive comments on our work examining the emergence of semantic role understanding in pre-trained decoder-only language models. We address each major comment below and indicate revisions to be made in the next version of the manuscript.","responses":[{"response":"We agree that linear probe accuracy improvements with scale do not by themselves isolate a shift toward more distributed representations, as they could alternatively reflect stronger or more redundant encodings. The manuscript's claim draws from the observed pattern of increasing linear decodability alongside the partial gap to fine-tuned performance, but we acknowledge the need for a more direct metric. In the revised manuscript we will qualify the statement in the abstract and §3 and add analyses of effective dimensionality of the probed subspaces (via participation ratio) together with linear-versus-nonlinear probe comparisons to better characterize any representational shift.","revision_made":"yes","referee_comment":"[Abstract / §3] Abstract and §3 (results on scale dependence): The claim that 'its internal implementation shifts toward more distributed representations as model scale increases' is not directly supported by the reported linear probe accuracies alone. Improved linear decodability with scale could reflect stronger, more redundant, or higher-dimensional encodings without necessarily indicating a change in distributedness; no explicit metric (e.g., effective dimensionality, sparsity measures, unit ablation, or linear-vs-nonlinear probe comparisons) is described to isolate this shift."},{"response":"We accept that explicit controls are necessary to support the inference of pre-existing encodings. The current manuscript relies on the standard linear-probe methodology and the gap between frozen and fine-tuned performance, but does not report the requested baselines. In the revised version we will add (i) random-label and random-feature baselines, (ii) statistical significance testing across multiple seeds, and (iii) a brief discussion of probe capacity relative to hidden dimension. These additions will clarify that the reported accuracies reflect information present in the frozen pre-trained representations.","revision_made":"yes","referee_comment":"[Abstract / Methods] Abstract and methods: The central inference that linear probe performance on frozen representations indicates semantic role information 'already encoded' during pre-training lacks reported controls for probe capacity, random baselines, or statistical significance testing. Without these, it remains possible that probes introduce or amplify structure rather than purely extract pre-existing encodings, weakening the emergence claim."}],"tokens_in":1350,"tokens_out":505,"duration_ms":43783,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main point is that linear probes recover semantic role labels from frozen decoder-only transformer representations, with accuracy rising across model scales yet staying below what fine-tuning achieves. This points to partial emergence of role structure from the pre-training objective alone. The paper does a clean job of using the frozen-versus-adapted comparison to quantify how much is already encoded versus what task-specific training adds, and focusing on semantic roles rather than more commonly probed syntactic features is a reasonable extension of existing work. The scale trend is worth noting for anyone thinking about how representations evolve with size. The softer spot is the interpretation that the internal implementation becomes more distributed as scale grows. Probe accuracy alone does not isolate that; it could simply reflect stronger or more redundant linear signals without changes in dimensionality, sparsity, or nonlinear dependence. The abstract gives no numbers, no baseline details, no probe-capacity controls, and no additional metrics like nonlinear probe comparisons or unit ablations, so the distributedness part rests on an inference that needs tighter evidence. This is the sort of incremental probing study that fits in the interpretability and emergence literature. Readers working on what LMs learn unsupervised or on reducing labeled data for semantic tasks would get some value from the setup, though they would probably want to see the full experimental controls before building on it. I would send it for peer review because the question is relevant and the basic design is straightforward; referees could push for the missing checks on the distributed claim without major rework.","headline":"Linear probes show semantic role info in frozen LM reps that improves with scale, but the claim of a shift to distributed representations lacks direct support.","tokens_in":2211,"tokens_out":365,"would_cite":false,"duration_ms":49624,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[],"headline":"Linear probing and neuron ablation on LM representations unrelated to RS distinction-forcing or J-cost","alignment":"orthogonal","rationale":"The paper's machinery (frozen linear probes on decoder-only transformers, CKA similarity, PCA/t-SNE clustering of semantic roles, role-conditioned ablation) studies empirical emergence of linguistic structure from next-token prediction. RS derives spacetime, constants, and J(x) = ½(x + x⁻¹) − 1 from a single distinction via theorems such as reality_from_one_distinction and absolute_floor_iff_bare_distinguishability. No shared structure (no cosh-cost, φ-ladder, 8-tick periodicity, or parameter-free constant derivation) appears; the domains are disjoint.","tokens_in":56701,"confidence":"high","tokens_out":167,"duration_ms":10254,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Semantic role understanding emerges in language models from pre-training, shifting to distributed representations at larger scales.","keywords":["semantic roles","language models","emergence","pre-training","linear probes","model scaling","decoder-only transformers","linguistic structure"],"falsifier":"Finding that semantic role information in large models is localized in specific layers or neurons, rather than distributed, or that small models show equally distributed encoding would challenge the claim of a scale-dependent shift.","tokens_in":2525,"feed_emoji":"🧠","tokens_out":511,"duration_ms":57462,"temperature":0.7,"pith_summary":"The authors examine if semantic role labeling, which captures who did what to whom, arises naturally from the language modeling task or requires additional training. They freeze pre-trained decoder-only transformers and train linear probes to predict semantic roles from the representations. Results show that substantial role information is already encoded during pre-training, with probe performance increasing alongside model size yet remaining below that of fully fine-tuned models. This points to partial emergence of linguistic structure from unsupervised objectives alone, with the encoding becoming more distributed as scale grows.","feed_headline":"Semantic roles emerge in pre-trained language models","feed_subtitle":"Frozen representations hold role information that spreads out more in bigger models","key_machinery":"Linear probes applied to frozen representations from decoder-only transformer language models, measuring the extractability of semantic role labels.","core_discovery":"Semantic role structure emerges from language modeling objectives, but its internal implementation shifts toward more distributed representations as model scale increases. By using linear probes on frozen models, the study reveals that pre-training encodes much of the necessary information for identifying semantic roles, although complete mastery still benefits from task-specific adaptation.","pith_inferences":["Other syntactic or semantic features might exhibit similar scale-dependent shifts in encoding style.","Interpretability techniques may need adaptation for very large models where information is highly distributed.","This suggests that scaling laws could apply not just to performance but to the geometry of learned linguistic structures."],"forward_implications":["Pre-training encodes substantial semantic role information without task-specific supervision.","Semantic role extraction accuracy improves with increasing model scale.","The internal representation of semantic roles becomes more distributed rather than localized in larger models.","Fine-tuning can still enhance performance beyond what pre-training provides alone."],"fun_headline_variants":["Pre-trained models encode semantic roles","Semantic role info emerges during pre-training","Role understanding partially arises in frozen LMs","Semantic roles distribute more in bigger models"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"That the accuracy of linear probes on frozen model layers directly measures the semantic role information present from pre-training without the probes creating new structure.","fun_headline_variants_meta":{"raw":{"variants":["Pre-trained models encode semantic roles","Semantic role info emerges during pre-training","Role understanding partially arises in frozen LMs","Semantic roles distribute more in bigger models"]},"model":"grok-4.3","cost_usd":0.003959,"raw_usage":{"total_tokens":1895,"prompt_tokens":569,"num_sources_used":0,"completion_tokens":49,"cost_in_usd_ticks":39590500,"prompt_tokens_details":{"text_tokens":569,"audio_tokens":0,"image_tokens":0,"cached_tokens":64},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1277,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":569,"tokens_out":49,"duration_ms":14938,"temperature":1.0,"reasoning_tokens":1277,"cache_read_input_tokens":64,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-12T01:50:16.320211+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Finding that semantic role information in large models is localized in specific layers or neurons, rather than distributed, or that small models show equally distributed encoding would challenge the claim of a scale-dependent shift.","supporting_citations":[],"review_version":1}