🧑‍🔬 Ting-Hao Kenneth Huang

Ting-Hao Kenneth Huang works on language technology that connects text with other kinds of information, such as images or tables, and on building datasets and evaluation tasks for it. Huang's two papers here share a method: asking people to write language for a defined task, then using that collection to test what systems can and cannot do. One paper addresses storytelling grounded in photos. The other addresses multi-step numerical reasoning over documents.

The most cited of these papers is FinQA (2021). It introduces a question-answering benchmark built on S&P 500 earnings reports. Finance professionals wrote the questions, and answering them often requires combining figures from tables with the surrounding prose and chaining several calculations. Each answer is paired with an executable reasoning program, so a system can be checked both on its final number and on the steps it took. The earlier Visual Storytelling (2016) moves image captioning from literal descriptions of single photos toward narratives about photo sequences. It presents the SIND v.1 dataset, which collects several tiers of text for the same images, and it tests automatic metrics against human judgments before choosing one.

Taken in order, the two papers shift from vision-and-language generation in 2016 to numerical reasoning over financial text and tables in 2021. Bylines from 2018 to 2025 list Pennsylvania State University, and those from 2011 to 2017 list Carnegie Mellon University. Co-authors on the papers here include Nasrin Mostafazadeh, Devi Parikh, Dhruv Batra, Wenhu Chen and Charese Smiley.

Ting-Hao Kenneth Huang has 2 papers in our library, in 2016–2021, most of them on Reasoning; often with Aishwarya Agrawal, C. Lawrence Zitnick, Charese Smiley.