While Abstract Meaning Representation (AMR) has historically improved performance on a range of NLP tasks, the benefit—or lack thereof—of AMR augmentation for modern LLMs is thus far unclear. In this paper, we attempt to reproduce recent work that reported substantial downstream gains from AMR augmentation, finding that these are likely due to specific choices in the experimental settings used: using a consistent and unified protocol for hyperparameter selection, we observe that text-only baselines consistently match or exceed the performance of AMR-augmented models. To investigate this null result, we introduce a perplexity-based probe measuring the degree to which AMR provides an LLM with supplemental relational knowledge not already available to the model. We find that AMR augmentation does not help LLMs improve their understanding of relational content in the sentence, indicating that augmenting these models with AMR offers no clear benefit on downstream tasks.

To guarantee that an LLM’s outputs conform to a specified structure, context-free grammar (CFG) decoding engines force the selection of next tokens to produce strings that conform to a given CFG. Current CFG-constrained decoding engines are highly optimized, but still suffer from the inherent costs arising from their massive per-step search space—i.e. the entire token vocabulary. This results in intractably high overhead for more complex CFGs, which is precisely the situation where CFG engines are most useful. In this paper, we introduce CFGzip, an offline technique for compressing the token search space, which massively reduces CFG engine overhead. In experiments, we report latency reduction of up to 75x during batched inference, cutting overhead down to ~1.2-2x on the hardest grammars: with CFGzip, constrained decoding is now possible at scale for complex CFGs.

The task of authorship style transfer involves rewriting text in the style of a target author while preserving the meaning of the original text. Existing style transfer methods train a single model on large corpora to model all target styles at once: this high-cost approach offers limited flexibility for target-specific adaptation, and often sacrifices meaning preservation for style transfer. In this paper, we propose AuthorMix: a lightweight, modular, and interpretable style transfer framework. We first train individual, style-specific LoRA adapters on a small set of high-resource authors: this allows for the rapid training of specialized adaptation models for each new target using layer-wise adapter mixing via reinforcement learning, necessitating only a handful of target- style training examples. AuthorMix ranks first on the combined style–meaning score among all baselines, including GPT-5.1, and substantially improves meaning preservation over the trained baselines; under human evaluation it is the only method best-or-tied on every dimension.

We make the case for language models over logical forms (LFLMs), arguing that such models are more data-efficient than their textual counterparts. To that end, we introduce the  ̲Graph-based  ̲Formal- ̲Logical  ̲Distributional  ̲Semantics (GFoLDS) prototype, a pretrained LM over graph representations of logical forms, as a proof-of-concept of LFLMs. Using GFoLDS, we present strong experimental evidence that LFLMs can leverage the built-in, basic linguistic knowledge inherent in such models to immediately begin learning more complex patterns. On downstream tasks, we show that GFoLDS vastly outperforms textual, transformer LMs (BERT) pretrained on the same data, indicating that LFLMs can learn with substantially less data than models over plain text. Furthermore, we show that the performance of this model is likely to scale with additional parameters and pretraining data, suggesting the viability of LFLMs in real-world applications.

Although the power of LLM tool-use agents has ignited a flurry of recent research in this area, the curation of tool-use training data remains an open problem—especially for online RL training. Existing approaches to synthetic tool-use data generation tend to be non-interactive and/or non-compositional. We introduce RandomWorld, a pipeline for the procedural generation of interactive tools and compositional tool-use data. We show that models tuned via SFT and RL on synthetic RandomWorld data improve on a range of tool-use benchmarks, and set the new SoTA for two metrics on the NESTFUL dataset. Further experiments show that downstream performance scales with the amount of RandomWorld-generated training data, opening up the possibility of further improvement through the use of entirely synthetic data.

In this paper, we propose a novel, automatic tree-based evaluation metric for LLM-generated step-by-step assembly instructions, that more accurately reflects spatiotemporal aspects of construction than traditional metrics such as BLEU and BERT similarity scores. We apply our proposed metric to the domain of sewing instructions, and show that our metric better correlates with manually-annotated error counts, demonstrating our metric’s superiority for evaluating the spatiotemporal soundness of sewing instructions. Further experiments show that our metric is more robust than traditional approaches against artificially-constructed counterfactual examples that are specifically constructed to confound metrics that rely on textual similarity.

Interaction between learner and feedback-giver has come into focus recently for post-training of Large Language Models (LLMs), through the use of reward models that judge the appropriateness of a model’s response. In this paper, we investigate whether Dialogue Games—goal-directed and rule-governed activities driven predominantly by verbal actions—can also serve as a source of feedback signals for learning.We introduce Playpen, an environment for off- and online learning through Dialogue Game self-play, and investigate a representative set of post-training methods: supervised fine-tuning; direct alignment (DPO); and reinforcement learning with Group Relative Policy Optimization (GRPO). We experiment with post-training a small LLM (Llama-3.1-8B-Instruct), evaluating performance on unseen instances of training games as well as unseen games, and on standard benchmarks. We find that imitation learning through SFT improves performance on unseen instances, but negatively impacts other skills, while interactive learning with GRPO shows balanced improvements without loss of skills. We release the framework and the baseline training setups to foster research in this promising new direction of “learning in (synthetic) interaction”.

NLI tasks necessitate a substantial degree of logical reasoning; as such, the remarkable performance of SoTA transformers on these tasks may lead us to believe that those models have learned to reason logically. The results presented in this paper demonstrate that (i) models fine-tuned on NLI datasets learn to treat external negation as a distractor, effectively ignoring its presence in hypothesis sentences; (ii) several near-SoTA encoder and encoder-decoder transformer models fail to inductively learn the law of the excluded middle for a single external negation prefix with respect to NLI tasks, despite extensive fine-tuning; (iii) those models which are are able to learn the law of the excluded middle for a single prefix are unable to generalize this pattern to similar prefixes. Given the critical role of negation in logical reasoning, we may conclude from these findings that transformers do not learn to reason logically when fine-tuned for NLI tasks. Furthermore, these results suggest that transformers may not be able to inductively learn the role of negation with respect to NLI tasks, calling into question their capacity to fully acquire logical reasoning abilities.

Modeling the most likely label when an annotation task is perspective-dependent discards relevant sources of variation that come from the annotators themselves. We present three approaches to modeling the controversiality of a particular text. First, we explicitly represented annotators using annotator embeddings to predict the training signals of each annotator’s selections in addition to a majority class label. This method leads to reduction in error relative to models without these features, allowing the overall result to influence the weights of each annotator on the final prediction. In a second set of experiments, annotators were not modeled individually but instead annotator judgments were combined in a pairwise fashion that allowed us to implicitly combine annotators. Overall, we found that aggregating and explicitly comparing annotators’ responses to a static document representation produced high-quality predictions in all datasets, though some systems struggle to account for large or variable numbers of annotators.