Tabular prediction, a well-established problem in machine learning, has consistently garnered significant research attention within academia and industry. Recently, with the rapid development of large language models (LLMs), there has been increasing exploration of how to apply LLMs to tabular prediction tasks. Many existing methods, however, typically rely on extensive pre-training or fine-tuning of LLMs, which demands considerable computational resources. To avoid this, we propose a retrieval-based approach that utilizes the powerful capabilities of LLMs in representation, comprehension, and inference. Our approach eliminates the need for training any modules or performing data augmentation, depending solely on information from target dataset. Experimental results reveal that, even without specialized training for tabular data, our method exhibits strong predictive performance on tabular prediction task, affirming its practicality and effectiveness.

Contrastive learning has emerged as a highly effective and versatile technique in information retrieval. However, its application within table retrieval remains limited, and often neglecting a typical phenomenon in table retrieval: one table can be associated with multiple, distinct queries. Directly applying traditional contrastive learning strategies may lead to semantic contrast conflicts during training, potentially impairing the quality of learned representations. Additionally, many current table retrieval methods still operate on query-table joint encoding, which introduces notable inefficiencies during both the training and retrieval processes. For this issue, this paper proposes ConTR, a tabular semantic contrastive learning method that simultaneously considers both inter-table and intra-table differences. By segmenting table into multiple vector representations and enabling the matching of diverse queries through differentiated vectors, thereby facilitating a more focused contrastive learning process. Retrieval experiments based on two typical table-related tasks validate the feasibility and effectiveness of proposed method.

Table-based fact verification (TFV) aims to classify whether a statement is entailed or refuted by a given table, which necessitates adept comprehension and logical inference skills among both tabular and textual data. The complexity of TFV is attributed to the involvement of both soft linguistic reasoning and hard symbolic reasoning. Existing studies tend to rely exclusively on table pre-trained models, lacking sufficient reasoning ability and treating various types of reasoning without distinction. In this paper, we propose a novel approach that transforms TFV task into a latent variable learning problem, employing a set of task-specific functions. Specifically, we leverage the hard Expectation-Maximization (EM) algorithm to ascertain the latent logic type underlying statements, then channel each statement through a specialized network designed for unique logical reasoning. Furthermore, we conduct a comprehensive exploration of the practical implementations of our proposed method in TFV task. Our approach diverges from the prevalent reliance on table-based pre-trained models, yet manages to surpass performance of various baseline models, exemplifying its efficacy and innovation.

Abstract Logical table‐to‐text generation is a task within the realm of natural language generation (NLG) that aims to generate coherent and logically faithful sentences based on tables. Unlike conventional NLG tasks, this task demands not only surface‐level fluency but also a high degree of logic‐level fidelity in the generated outputs. Current table‐to‐text systems grapple with various quality issues, such as repetitive generation, insufficient reasoning and limited complexity. Therefore, we introduce LogicMoE, a dedicated Mixture‐of‐Experts (MoE) model tailored for logical table‐to‐text generation. The primary objective of LogicMoE is to enrich the diversity of generated sentences from both semantic and logical perspectives. In particular, each expert within the model serves as a specialized generator responsible for generating sentences of a specific logical type. Additionally, we propose and employ novel evaluation metrics to comprehensively assess the diversity of generated outputs. Our experimental results showcase LogicMoE's superiority with absolute improvements of 0.8 and 2.2 in BLEU‐3 over the strong baselines on LogicNLG and Logic2Text datasets, respectively, driving the state‐of‐the‐art performance to a new level. Furthermore, we highlight its inherent advantages in terms of diversity and controllability, signifying its potential to spearhead advancements in logical table‐to‐text generation applications.