HeLM: Highlighted Evidence augmented Language Model for Enhanced Table-to-Text Generation

Anonymous

TL;DR: LLaMA2 based table2text generation framework comprises a table reasoner and a table summarizer.

Abstract: To harness the powerful text generation capabilities of recent large language models (LLMs) in the Table-to-Text task, employing parameter-efficient fine-tuning on open-source LLMs is a viable approach. However, how to enhance the model's table reasoning ability during LLM fine-tuning presents a challenge. In this study, we propose a two-step solution called HeLM. Different from previous fine-tuning-based methods that directly expand tables as inputs, our approach injects reasoning information into the input table by emphasizing table-specific row data. Our model consists of two modules: a table reasoner that identifies relevant row evidence, and a table summarizer that generates sentences based on the highlighted table. To facilitate this, we propose a method to train and construct reasoning labels for obtaining the table reasoner. On both the FetaQA and QTSumm datasets, our approach achieved state-of-the-art results in ROUGE and BLEU scores. Additionally, it is observed that highlighting input tables significantly enhances the model's performance and provides valuable interpretability.

Paper Type: long

Research Area: Generation

Contribution Types: Model analysis & interpretability

Languages Studied: English