An IDR Framework of Opportunities and Barriers between HCI and NLP

Nanna Inie, Leon DerczynskiIT 2021

TL;DR

This paper proposes a framework to bridge the gap between Human-Computer Interaction (HCI) and Natural Language Processing (NLP). It identifies opportunities for cross-pollination and potential barriers, aiming to inspire more interdisciplinary research and collaboration between these two fields.

What Problem Does This Paper Address?

Despite the clear potential for mutual benefit, research at the intersection of HCI and NLP remains relatively rare. This gap is attributed to various factors, including differing methodologies and evaluation approaches, the perception that NLP isn't advanced enough for HCI needs, and the technical skills required to implement NLP systems. Furthermore, HCI is sometimes viewed as a "soft science" lacking the quantitative rigor of computer science. This paper aims to address this by providing a structured way to understand the landscape of opportunities and challenges when combining these fields.

Methods

The paper's core contribution is the development of an Interdisciplinary Research (IDR) framework. This framework is built by combining knowledge from both HCI and NLP, following a model that identifies research topics, concepts, disciplines, theories, and terminology from each field to create an integrated perspective. The authors' method involved debating and discussing research processes and methods, highlighting differences and similarities between HCI and NLP. The framework is presented in a table and elaborated upon in the text, categorizing opportunities and barriers across methods/analyses and outputs.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Mutual Benefit is Underrealized: Both HCI and NLP have much to gain from each other's methods, analyses, and findings, yet this potential is not being fully exploited.
  • HCI's Strengths for NLP:
    • Ethnographic Methods: Can provide deeper context and understanding of why and how language is used, leading to more effective NLP system deployment.
  • Annotation Process Insights: HCI can help improve NLP dataset creation by focusing on the situated understanding of phenomena and the human aspect of annotation.
  • Qualitative Error Analysis: HCI's rich methodologies can offer new ways to understand and analyze NLP model failures beyond purely quantitative metrics.
  • New Abstractions: HCI concepts can provide novel ways to think about and structure NLP technology (e.g., "design material").
  • Extra Data Layers: HCI generates vast amounts of transcribed data (interviews, field notes) that can serve as valuable resources for NLP training.
  • NLP's Strengths for HCI:
    • Automatic Transcription: Can significantly speed up the processing of qualitative data in HCI, though current tools are often limited.
  • Thematic Classification/Clustering: NLP can help identify themes in qualitative data more systematically and at scale, potentially revealing patterns missed by human coders.
  • Novel Analysis Tools: NLP offers powerful tools for analyzing large text datasets, uncovering new insights in HCI research.
  • Barriers are Significant: Technical barriers (coding skills, limited modalities for NLP) and methodological differences (detaching from quantitative methods, terminology mismatches) pose significant challenges.
  • Output Risks: Biases in data are a major concern, as understanding biases from another discipline is difficult, potentially leading to misapplication or flawed assumptions.

Key Results

The paper doesn't present empirical results from a specific experiment. Instead, its key "result" is the framework itself, which categorizes:

  • Opportunities for NLP: Ethnographic methods, annotation processes, error analysis, new abstractions, extra data layers.
  • Opportunities for HCI: Automatic transcription, thematic classification/clustering, novel analysis tools.
  • Barriers/Risks (Human): Coding skills, detaching from quantitative methods, terminology, critical thinking.
  • Barriers/Risks (Technical): Limited modalities (speech & text processing), data access.
  • Output Opportunities: Discoveries, quality analyses, artwork, network.
  • Output Risks: Biases in data.

The paper provides numerous examples of how these opportunities have been or could be realized, citing existing research.

Limitations and Future Work

The authors acknowledge that their framework is a "brief selection of topics" and that it "could and should be expanded in use by researchers from both fields." They invite researchers to expand upon the categories, both in detail and quantity. Future work would involve applying this framework to guide specific interdisciplinary projects, further refining the categories, and developing concrete strategies to overcome the identified barriers. The paper also implicitly suggests the need for better tools and educational resources to lower the technical barrier for interdisciplinary work.

Summary updated

Abstract

This paper presents a framework of opportunities and barriers/risks between the two research fields Natural Language Processing (NLP) and Human-Computer Interaction (HCI). The framework is constructed by following an interdisciplinary research-model (IDR), combining field-specific knowledge with existing work in the two fields. The resulting framework is intended as a departure point for discussion and inspiration for research collaborations.
BibTeX
@inproceedings{inie2021idr,
  author = {Nanna Inie and Leon Derczynski},
  title = {{An IDR Framework of Opportunities and Barriers between HCI and NLP}},
  booktitle = {IT},
  year = {2021},
}
Figure