Xiaochang Peng

Aaron Chan, Shaoliang Nie, Liang Tan et al. · meta-ai

Following how humans communicate, free-text rationales aim to use natural language to explain neural language model (LM) behavior. However, free-text rationales' unconstrained nature makes them prone to hallucination, so it is important to have metrics for free-text rationale quality. Existing free-text rationale metrics measure how consistent the rationale is with the LM's predicted label, but there is no protocol for assessing such metrics' reliability. Thus, we propose FRAME, a framework for evaluating rationale-label consistency (RLC) metrics for free-text rationales. FRAME is based on three axioms: (1) good metrics should yield highest scores for reference rationales, which maximize RLC by construction; (2) good metrics should be appropriately sensitive to semantic perturbation of rationales; and (3) good metrics should be robust to variation in the LM's task performance. Across three text classification datasets, we show that existing RLC metrics cannot satisfy all three FRAME axioms, since they are implemented via model pretraining which muddles the metric's signal. Then, we introduce a non-pretraining RLC metric that greatly outperforms baselines on (1) and (3), while performing competitively on (2). Finally, we discuss the limitations of using RLC to evaluate free-text rationales.

Hamed Firooz, Rui Liu, Yuchen Lu et al.

Content moderation at scale remains one of the most pressing challenges in today's digital ecosystem, where billions of user- and AI-generated artifacts must be continuously evaluated for policy violations. Although recent advances in large language models (LLMs) have demonstrated strong potential for policy-grounded moderation, the practical challenges of training these systems to achieve expert-level accuracy in real-world settings remain largely unexplored, particularly in regimes characterized by label sparsity, evolving policy definitions, and the need for nuanced reasoning beyond shallow pattern matching. In this work, we present a comprehensive empirical investigation of scaling reinforcement learning (RL) for content classification, systematically evaluating multiple RL training recipes and reward-shaping strategies-including verifiable rewards and LLM-as-judge frameworks-to transform general-purpose language models into specialized, policy-aligned classifiers across three real-world content moderation tasks. Our findings provide actionable insights for industrial-scale moderation systems, demonstrating that RL exhibits sigmoid-like scaling behavior in which performance improves smoothly with increased training data, rollouts, and optimization steps before gradually saturating. Moreover, we show that RL substantially improves performance on tasks requiring complex policy-grounded reasoning while achieving up to 100x higher data efficiency than supervised fine-tuning, making it particularly effective in domains where expert annotations are scarce or costly.

Aaron Chan, Maziar Sanjabi, Lambert Mathias et al.

An extractive rationale explains a language model's (LM's) prediction on a given task instance by highlighting the text inputs that most influenced the prediction. Ideally, rationale extraction should be faithful (reflective of LM's actual behavior) and plausible (convincing to humans), without compromising the LM's (i.e., task model's) task performance. Although attribution algorithms and select-predict pipelines are commonly used in rationale extraction, they both rely on certain heuristics that hinder them from satisfying all three desiderata. In light of this, we propose UNIREX, a flexible learning framework that generalizes rationale extractor optimization as follows: (1) specify architecture for a learned rationale extractor; (2) select explainability objectives (i.e., faithfulness and plausibility criteria); and (3) jointly the train task model and rationale extractor on the task using the selected objectives. UNIREX enables replacing prior works' heuristic design choices with a generic learned rationale extractor in (1) and optimizing it for all three desiderata in (2)-(3). To facilitate comparison between methods with respect to multiple desiderata, we introduce the Normalized Relative Gain (NRG) metric. Across five text classification datasets, our best UNIREX configuration outperforms baselines by an average of 32.9% NRG. Plus, we find that UNIREX-trained rationale extractors can even generalize to unseen datasets and tasks.

Xiaochang Peng, Chuan Wang, Daniel Gildea et al.

Neural attention models have achieved great success in different NLP tasks. How- ever, they have not fulfilled their promise on the AMR parsing task due to the data sparsity issue. In this paper, we de- scribe a sequence-to-sequence model for AMR parsing and present different ways to tackle the data sparsity problem. We show that our methods achieve significant improvement over a baseline neural atten- tion model and our results are also compet- itive against state-of-the-art systems that do not use extra linguistic resources.

Linfeng Song, Xiaochang Peng, Yue Zhang et al.

This paper addresses the task of AMR-to-text generation by leveraging synchronous node replacement grammar. During training, graph-to-string rules are learned using a heuristic extraction algorithm. At test time, a graph transducer is applied to collapse input AMRs and generate output sentences. Evaluated on SemEval-2016 Task 8, our method gives a BLEU score of 25.62, which is the best reported so far.

Linfeng Song, Yue Zhang, Xiaochang Peng et al.

The task of AMR-to-text generation is to generate grammatical text that sustains the semantic meaning for a given AMR graph. We at- tack the task by first partitioning the AMR graph into smaller fragments, and then generating the translation for each fragment, before finally deciding the order by solving an asymmetric generalized traveling salesman problem (AGTSP). A Maximum Entropy classifier is trained to estimate the traveling costs, and a TSP solver is used to find the optimized solution. The final model reports a BLEU score of 22.44 on the SemEval-2016 Task8 dataset.

Xiaochang Peng, Daniel Gildea

In this paper, we introduce a variation of the skip-gram model which jointly learns distributed word vector representations and their way of composing to form phrase embeddings. In particular, we propose a learning procedure that incorporates a phrase-compositionality function which can capture how we want to compose phrases vectors from their component word vectors. Our experiments show improvement in word and phrase similarity tasks as well as syntactic tasks like dependency parsing using the proposed joint models.