{"id":"994d3c7c-3385-40ec-b898-6e846686e59b","arxiv_id":"2606.10059","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"A new worsening-trick construction compiles arbitrary-context rewrite rules A → B / L _ R into FSTs with short uniform formulas that match prior transducers where semantics coincide.","lead":"The paper introduces a compilation method called the worsening trick that generates rewrite candidates and filters worse ones to produce finite-state transducers for general rewrite rules. A smart generalist might read it because it offers a simpler way to implement string rewriting used in speech, text processing, and language technology.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"Reader's weakest assumption was extracted from the abstract alone. The full text supplies concrete validation evidence that directly exercises the filtering step on the relevant cases, removing the load-bearing status of the assumption for the reported scope.","tokens_in":1738,"tokens_out":208,"duration_ms":9583,"concrete_test":"Re-run the full regression suite after adding 20 new overlapping-match test cases with nested contexts and directed weights; verify that every generated transducer is identical to the foma reference (modulo state renumbering).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The manuscript reports that the worsening-trick transducers match foma exactly (apart from state numbering) on a substantial collection of rewrite grammars plus an automated regression suite covering the major modalities, including overlapping matches and multiple contexts. This empirical check directly tests whether the total order on 'worse' candidates reproduces the intended semantics without omissions or spurious rewrites.","agreement_with_reader":"disagree"},"referee_report":{"model":"grok-4.3","summary":"The paper introduces the 'worsening trick' for compiling general rewrite rules A \to B / L _ R (with A, B, L, R regular) into finite-state transducers. The method generates all legal rewrite candidates for an input and filters those that are worse than another candidate under a total order; the resulting transducers are claimed to be short and uniform, to support multiple contexts, arbitrary transductions, markup, directed rewriting, weights, and parallel rewriting, and to reproduce the transducers of Kaplan-Kay and Karttunen constructions wherever the semantics coincide. The construction is implemented in PyFoma and validated by exact transducer match (apart from state numbering) against foma on a substantial grammar collection plus an automated regression suite covering overlapping matches and multiple contexts.","tokens_in":1811,"tokens_out":568,"duration_ms":11697,"significance":"If the central filtering step is correct, the approach supplies a compact, uniform algorithmic procedure that avoids intricate marker compositions and is easier to extend to the listed modalities. The external validation against foma on a broad test suite directly addresses the correctness of the total order for the supported cases and constitutes a reproducible empirical check.","major_comments":[{"comment":"§4 (Worsening Trick construction): the total order used to define 'worse' candidates is described algorithmically but lacks an explicit inductive argument or edge-case enumeration showing that it never omits a rewrite or introduces a spurious one for overlapping matches; the empirical match with foma is strong but does not substitute for such an argument when the claim is that the method is semantically equivalent wherever prior constructions apply.","section":"§4"},{"comment":"§5 (Validation): the regression suite is reported to cover 'major modalities' including overlapping matches, yet the paper does not list the precise set of test grammars or the coverage metric for context-interaction cases; without this, it is difficult to assess whether the validation fully exercises the weakest assumption identified in the construction.","section":"§5"}],"minor_comments":[{"comment":"The abstract states that formulas are 'short and uniform' but does not exhibit the concrete transducer expressions; placing one or two representative formulas (with the worsening filter) in the abstract or introduction would make the central contribution immediately visible.","section":"Abstract"},{"comment":"Notation for the total order on candidates is introduced without a dedicated definition environment; a numbered definition would improve readability when the order is referenced in later sections.","section":"§4"},{"comment":"The paper mentions support for weights and parallel rewriting but does not show how the worsening order interacts with weighted transducers; a brief remark or example would clarify the extension.","section":"§6"}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the careful reading, positive assessment, and constructive comments on our manuscript. We address the major comments point by point below.","responses":[{"response":"We agree that an explicit inductive argument and edge-case enumeration for overlapping matches would strengthen the theoretical justification of the total order. The manuscript currently emphasizes the empirical equivalence with foma (exact match apart from state numbering) on a broad suite as evidence of correctness where semantics coincide. In revision we will add a dedicated subsection providing an inductive argument that the filtering step neither omits valid rewrites nor introduces spurious ones, together with explicit enumeration of the principal overlapping-match cases.","revision_made":"yes","referee_comment":"[§4] §4 (Worsening Trick construction): the total order used to define 'worse' candidates is described algorithmically but lacks an explicit inductive argument or edge-case enumeration showing that it never omits a rewrite or introduces a spurious one for overlapping matches; the empirical match with foma is strong but does not substitute for such an argument when the claim is that the method is semantically equivalent wherever prior constructions apply."},{"response":"We accept that greater specificity on the regression suite would improve reproducibility and allow readers to evaluate coverage of context-interaction cases. In the revised version we will append the complete list of test grammars used in the automated regression suite and report the coverage metrics for overlapping matches and multiple-context interactions.","revision_made":"yes","referee_comment":"[§5] §5 (Validation): the regression suite is reported to cover 'major modalities' including overlapping matches, yet the paper does not list the precise set of test grammars or the coverage metric for context-interaction cases; without this, it is difficult to assess whether the validation fully exercises the weakest assumption identified in the construction."}],"tokens_in":1431,"tokens_out":401,"duration_ms":19545,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main point is a new compilation method for general rewrite rules A \to B / L _ R that generates candidate transducers and filters them by a total order on 'worse' outputs. This produces short, uniform formulas without the auxiliary markers used in Kaplan-Kay or Karttunen constructions.\n\nThe paper does a few things well. It shows the scheme handles multiple contexts, arbitrary transductions, markup, directed rewriting, weights, and parallel rules in one framework. The PyFoma implementation was checked against foma on a substantial collection of grammars plus an automated regression suite that covers overlapping matches and multiple contexts. The transducers match exactly apart from state numbering, which directly tests whether the filtering step reproduces the intended semantics.\n\nThe empirical validation is the strongest part. It addresses the obvious worry about whether the total order on worse candidates misses or adds rewrites in edge cases. With the regression suite covering the major modalities, the central claim holds up in practice.\n\nA minor soft spot is that the work stays at the level of an algorithmic procedure and external checks rather than supplying full derivations or machine-checked proofs for every overlap scenario. That is not a load-bearing flaw given the test coverage, but readers who want formal guarantees might still ask for it.\n\nThis is for people who build or maintain finite-state toolkits in computational linguistics. It is worth a serious referee because the validation is external and the construction is presented as a genuine alternative rather than a restatement.","headline":"The worsening trick gives a compact marker-free way to compile rewrite rules to FSTs that matches foma on tested cases including overlaps.","tokens_in":2280,"tokens_out":372,"would_cite":true,"duration_ms":14680,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"The worsening trick compiles general rewrite rules into finite-state transducers using short and uniform formulas.","keywords":["finite-state transducers","rewrite rules","worsening trick","transducer compilation","regular languages","computational linguistics","phonological rules","morphological rules"],"falsifier":"A comparison of the input-output mappings accepted by the worsening-based transducer versus a marker-based transducer on a rule with overlapping possible match positions would show a difference if the method fails.","tokens_in":2633,"feed_emoji":"🔄","tokens_out":570,"duration_ms":22090,"temperature":0.7,"pith_summary":"The paper establishes a compilation method for rewrite rules of the form A to B in left and right contexts. It relies on generating candidate rewrites for each input and then removing those that are worse than some other candidate according to a total order. This produces transducers that handle overlapping matches, multiple contexts, weights, and parallel application. The formulas are claimed to be shorter and easier to extend than previous constructions while producing equivalent results where the semantics match. Readers in computational linguistics would care because this simplifies the creation of models for phonological and morphological processes.","feed_headline":"Worsening trick yields short formulas for rewrite rule transducers","feed_subtitle":"It generates and filters candidates to handle contexts and weights while matching prior results.","key_machinery":"The worsening trick, which filters rewrite candidates using a total order on 'worse' to select the correct outputs.","core_discovery":"By generating all legal rewrite candidates and filtering out those that are worse than another candidate for the same input, the worsening trick yields compact transducers for general rewrite rules that correctly implement the intended semantics for arbitrary regular languages in A, B, L, and R.","pith_inferences":["This approach may allow for more straightforward extensions to new rewrite modalities without complex adjustments to compositions.","The uniform nature of the formulas could facilitate formal proofs of correctness for rewrite semantics.","Applications in natural language processing might benefit from easier integration of weighted rewrite rules."],"forward_implications":["The construction supports multiple contexts, arbitrary transductions, markup, directed rewriting, weights, and parallel rewriting.","The resulting formulas are short and uniform.","The transducers match those produced by earlier marker-based methods where the semantics coincide.","The method has been validated to produce matching results on collections of rewrite grammars and regression tests covering major modalities."],"fun_headline_variants":["Worsening trick compiles rewrite rules to FSTs","Short formulas for rewrite rule FSTs using worsening trick","Rewrite rules to finite-state transducers with worsening trick","Worsening trick for compact rewrite rule transducers"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"Filtering candidates by a total order on which is worse correctly implements the rewrite semantics without introducing or omitting rewrites in overlapping or context-sensitive cases.","fun_headline_variants_meta":{"raw":{"variants":["Worsening trick compiles rewrite rules to FSTs","Short formulas for rewrite rule FSTs using worsening trick","Rewrite rules to finite-state transducers with worsening trick","Worsening trick for compact rewrite rule transducers"]},"model":"grok-4.3","cost_usd":0.007549,"raw_usage":{"total_tokens":3445,"prompt_tokens":636,"num_sources_used":0,"completion_tokens":61,"cost_in_usd_ticks":75487000,"prompt_tokens_details":{"text_tokens":636,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2748,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":636,"tokens_out":61,"duration_ms":15828,"temperature":1.0,"reasoning_tokens":2748,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-27T13:53:08.627597+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A comparison of the input-output mappings accepted by the worsening-based transducer versus a marker-based transducer on a rule with overlapping possible match positions would show a difference if the method fails.","supporting_citations":[],"review_version":1}