Recent advances in Language Models (LMs) have failed to mask their shortcomings particularly in the domain of reasoning. This limitation impacts several tasks, most notably those involving ontology engineering. As part of a PhD research, we investigate the consequences of incorporating formal methods on the performance of Small Language Models (SLMs) on reasoning tasks. Specifically, we aim to orient our work toward using SLMs to bootstrap ontology construction and set up a series of preliminary experiments to determine the impact of expressing logical problems with different grammars on the performance of SLMs on a predefined reasoning task. Our findings show that it is possible to substitute Natural Language (NL) with a more compact logical language while maintaining a strong performance on reasoning tasks and hope to use these results to further refine the role of SLMs in ontology engineering.

Model merging is an efficient way of obtaining a multi-task model from several pretrained models without further fine-tuning, and it has gained attention in various domains, including natural language processing (NLP). Despite the efficiency, a key challenge in model merging is the seemingly inevitable decrease in task performance as the number of models increases. In this paper, we propose $\mathbf{S}$pectral $\mathbf{T}$runcation $\mathbf{A}$nd $\mathbf{R}$escale (STAR) that aims at mitigating ``merging conflicts'' by truncating small components in the respective spectral spaces, which is followed by an automatic parameter rescaling scheme to retain the nuclear norm of the original matrix. STAR requires no additional inference on original training data and is robust to hyperparamater choice. We demonstrate the effectiveness of STAR through extensive model merging cases on diverse NLP tasks. Specifically, STAR works robustly across varying model sizes, and can outperform baselines by 4.2$\%$ when merging 12 models on Flan-T5. Our code is publicly available at https://github.com/IBM/STAR.

Task-oriented dialogue systems are essential for applications ranging from customer service to personal assistants and are widely used across various industries. However, developing effective multi-domain systems remains a significant challenge due to the complexity of handling diverse user intents, entity types, and domain-specific knowledge across several domains. In this work, we propose DARD (Domain Assigned Response Delegation), a multi-agent conversational system capable of successfully handling multi-domain dialogs. DARD leverages domain-specific agents, orchestrated by a central dialog manager agent. Our extensive experiments compare and utilize various agent modeling approaches, combining the strengths of smaller fine-tuned models (Flan-T5-large & Mistral-7B) with their larger counterparts, Large Language Models (LLMs) (Claude Sonnet 3.0). We provide insights into the strengths and limitations of each approach, highlighting the benefits of our multi-agent framework in terms of flexibility and composability. We evaluate DARD using the well-established MultiWOZ benchmark, achieving state-of-the-art performance by improving the dialogue inform rate by 6.6% and the success rate by 4.1% over the best-performing existing approaches. Additionally, we discuss various annotator discrepancies and issues within the MultiWOZ dataset and its evaluation system.

A major limitation of prompt tuning is its dependence on large labeled training datasets. Under few-shot learning settings, prompt tuning lags far behind full-model fine-tuning, limiting its scope of application. In this paper, we leverage the powerful LLMs to synthesize task-specific labeled data for training the soft prompts. We first introduce a distribution-aligned weighted generator tuning (DawGen) method to encourage generating in-distribution data that aligns with the few-shot real data. Then, we train soft prompts on both synthetic and real datasets using a gradient surgery approach, which eliminates the conflicting gradients from different data sources. Experiments on seven sentence-pair classification datasets demonstrate the effectiveness of our proposed method for boosting prompt tuning in few-shot learning settings. Results on QQP, MRPC, and SICK datasets are even comparable to the performance of transfer learning from large real-world datasets, showing the promise of synthetic data as an alternative for enhancing soft prompt tuning.

The inherent ambiguity of cause and effect boundaries poses a challenge in evaluating causal event extraction tasks. Traditional metrics like Exact Match and BertScore poorly reflect model performance, so we trained evaluation models to approximate human evaluation, achieving high agreement. We used them to perform Reinforcement Learning with extraction models to align them with human preference, prioritising semantic understanding. We successfully explored our approach through multiple datasets, including transferring an evaluator trained on one dataset to another as a way to decrease the reliance on human-annotated data. In that vein, we also propose a weak-to-strong supervision method that uses a fraction of the annotated data to train an evaluation model while still achieving high performance in training an RL model. Our code is available at https://github.com/oyarsa/event_extraction/tree/causal-event-extraction.

We study the problem of automatically annotating relevant numerals (GAAP metrics) occurring in the financial documents with their corresponding XBRL tags. Different from prior works, we investigate the feasibility of solving this extreme classification problem using a generative paradigm through instruction tuning of Large Language Models (LLMs). To this end, we leverage metric metadata information to frame our target outputs while proposing a parameter efficient solution for the task using LoRA. We perform experiments on two recently released financial numeric labeling datasets. Our proposed model, FLAN-FinXC, achieves new state-of-the-art performances on both the datasets, outperforming several strong baselines. We explain the better scores of our proposed model by demonstrating its capability for zero-shot as well as the least frequently occurring tags. Also, even when we fail to predict the XBRL tags correctly, our generated output has substantial overlap with the ground-truth in majority of the cases.

This paper describes our submission to Task 2 of SemEval-2024: Safe Biomedical Natural Language Inference for Clinical Trials. The Multi-evidence Natural Language Inference for Clinical Trial Data (NLI4CT) consists of a Textual Entailment (TE) task focused on the evaluation of the consistency and faithfulness of Natural Language Inference (NLI) models applied to Clinical Trial Reports (CTR). We test 2 distinct approaches, one based on finetuning and ensembling Masked Language Models and the other based on prompting Large Language Models using templates, in particular, using Chain-Of-Thought and Contrastive Chain-Of-Thought. Prompting Flan-T5-large in a 2-shot setting leads to our best system that achieves 0.57 F1 score, 0.64 Faithfulness, and 0.56 Consistency.

Large Language Models (LLMs) have demonstrated impressive capabilities to solve a wide range of tasks without being explicitly fine-tuned on task-specific datasets. However, deploying LLMs in the real world is not trivial, as it requires substantial computing resources. In this paper, we investigate whether smaller, compact LLMs are a good alternative to the comparatively Larger LLMs2 to address significant costs associated with utilizing LLMs in the real world. In this regard, we study the meeting summarization task in a real-world industrial environment and conduct extensive experiments by comparing the performance of fine-tuned compact LLMs (e.g., FLAN-T5, TinyLLaMA, LiteLLaMA) with zero-shot larger LLMs (e.g., LLaMA-2, GPT-3.5, PaLM-2). We observe that most smaller LLMs, even after fine-tuning, fail to outperform larger zero-shot LLMs in meeting summarization datasets. However, a notable exception is FLAN-T5 (780M parameters), which performs on par or even better than many zero-shot Larger LLMs (from 7B to above 70B parameters), while being significantly smaller. This makes compact LLMs like FLAN-T5 a suitable cost-efficient solution for real-world industrial deployment.

Modular and composable transfer learning is an emerging direction in the field of Parameter Efficient Fine-Tuning, as it enables neural networks to better organize various aspects of knowledge, leading to improved cross-task generalization. In this paper, we introduce a novel approach Customized Polytropon (C-Poly) that combines task-common skills and task-specific skills, while the skill parameters being highly parameterized using low-rank techniques. Each task is associated with a customizable number of exclusive specialized skills and also benefits from skills shared with peer tasks. A skill assignment matrix is jointly learned. To evaluate our approach, we conducted extensive experiments on the Super-NaturalInstructions and the SuperGLUE benchmarks. Our findings demonstrate that C-Poly outperforms fully-shared, task-specific, and skill-indistinguishable baselines, significantly enhancing the sample efficiency in multi-task learning scenarios. As the number of parameters in Large Language Models (LLMs) continues to grow, training these models efficiently with limited computational resources has become a challenge. In recent years, there has been a shift towards employing Parameter Effective Fine-Tuning (PEFT) methods to address this issue. Examples of such methods include LoRA (Hu et al., 2022), AdaLoRA (Zhang et al., 2023a), and (IA) These methods focus on fine-tuning the adapter while freezing the pre-trained model, effectively reducing the computational cost.

The derivation of mathematical results in specialised fields, using Large Language Models (LLMs), is an emerging research direction that can help identify models' limitations, and potentially support mathematical discovery. In this paper, we leverage a symbolic engine to generate derivations of equations at scale, and investigate the capabilities of LLMs when deriving goal equations from premises. Specifically, we employ in-context learning for GPT and fine-tune a range of T5 models to compare the robustness and generalisation of pre-training strategies to specialised models. Empirical results show that fine-tuned FLAN-T5-large (MathT5) outperforms GPT models on all static and out-of-distribution test sets in conventional scores. However, an in-depth analysis reveals that the fine-tuned models are more sensitive to perturbations involving unseen symbols and (to a lesser extent) changes to equation structure. In addition, we analyse 1.7K equations, and over 200 derivations, to highlight common reasoning errors such as the inclusion of incorrect, irrelevant, and redundant equations. Finally, we explore the suitability of existing metrics for evaluating mathematical derivations and find evidence that, while they can capture general properties such as sensitivity to perturbations, they fail to highlight fine-grained reasoning errors and essential differences between models. Overall, this work demonstrates that training models on synthetic data may improve their math capabilities beyond much larger LLMs, but current metrics are not appropriately assessing the quality of generated mathematical text.