We introduce SWAN (Semantic Watermarking with Abstract Meaning Representation), a novel framework that embeds watermark signatures into the semantic structure of a sentence using Abstract Meaning Representation (AMR). In contrast to existing watermarking methods, which typically encode signatures by adjusting token selection preferences during text generation, SWAN embeds the signature directly in the sentence’s semantic representation. As the signature is encoded at the semantic structure level, any paraphrase that preserves meaning, automatically preserves the signature. SWAN is training-free: watermark injection is achieved by prompting an LLM to generate sentences guided by a selected AMR template while maintaining contextual coherence, and detection uses an off-the-shelf AMR parser followed by a simple one-proportion z-test. Empirical evaluation on the RealNews benchmark shows SWAN matches state-of-the-art detection performance on unaltered watermarked text, while significantly improving robustness against paraphrasing, increasing detection AUC by up to 13.9 percentage points compared to prior methods. These results demonstrate that SWAN’s approach of anchoring watermarks in AMR semantic structures provides a simple, effective, and prompt-based method for robust text provenance verification under paraphrasing, opening new avenues for semantic-level watermarking research.

Responsible use of Authorship Verification (AV) systems not only requires high accuracy but also interpretable solutions. More importantly, for systems to be used to make decisions with real-world consequences requires the model’s prediction to be explainable using interpretable features that can be traced to the original texts. Neural methods achieve high accuracies, but their representations lack direct interpretability. Furthermore, LLM predictions cannot be explained faithfully – if there is an explanation given for a prediction, it doesn’t represent the reasoning process behind the model’s prediction. In this paper, we introduce Residualized Similarity (RS), a novel method that supplements systems using interpretable features with a neural network to improve their performance while maintaining interpretability. Authorship verification is fundamentally a similarity task, where the goal is to measure how alike two documents are. The key idea is to use the neural network to predict a similarity residual, i.e. the error in the similarity predicted by the interpretable system. Our evaluation across four datasets shows that not only can we match the performance of state-of-the-art authorship verification models, but we can show how and to what degree the final prediction is faithful and interpretable.

Language understanding includes identifying logical connections between events in a discourse, such as news and instructional text. We study the transferability of causal knowledge across these two domains by analyzing the extent to which understanding preconditions in narratives such as news articles can help models reason about cooking recipes, and vice-versa. Our experiments show that using instructions to pretrain small models on one domain before similarly finetuning it on the other shows a slight improvement over just finetuning it. We also find that finetuning the models on a mix of both types of data is better (~3-7%) for understanding causal relations in instructional text. While we find that the improvements do not translate to larger or already instruction tuned models, our analysis highlights the aspects of a plan that are better captured through the interoperability of causal knowledge.

Social science NLP tasks, such as emotion or humor detection, are required to capture the semantics along with the implicit pragmatics from text, often with limited amounts of training data. Instruction tuning has been shown to improve the many capabilities of large language models (LLMs) such as commonsense reasoning, reading comprehension, and computer programming. However, little is known about the effectiveness of instruction tuning on the social domain where implicit pragmatic cues are often needed to be captured. We explore the use of instruction tuning for social science NLP tasks and introduce Socialite-Llama — an open-source, instruction-tuned Llama. On a suite of 20 social science tasks, Socialite-Llama improves upon the performance of Llama as well as matches or improves upon the performance of a state-of-the-art, multi-task finetuned model on a majority of them. Further, Socialite-Llama also leads to improvement on 5 out of 6 related social tasks as compared to Llama, suggesting instruction tuning can lead to generalized social understanding. All resources including our code, model and dataset can be found through [bit.ly/socialitellama](https://bit.ly/socialitellama/).

Research on psychological risk factors for suicide has developed for decades. However, combining explainable theory with modern data-driven language model approaches is non-trivial. In this study, we propose and evaluate methods for identifying language patterns aligned with theories of suicide risk by combining theory-driven suicidal archetypes with language model-based and relative entropy-based approaches. Archetypes are based on prototypical statements that evince risk of suicidality while relative entropy considers the ratio of how unusual both a risk-familiar and unfamiliar model find the statements. While both approaches independently performed similarly, we find that combining the two significantly improved the performance in the shared task evaluations, yielding our combined system submission with a BERTScore Recall of 0.906. Consistent with the literature, we find that titles are highly informative as suicide risk evidence, despite the brevity. We conclude that a combination of theory- and data-driven methods are needed in the mental health space and can outperform more modern prompt-based methods.