The IBM natural conversation framework: a new paradigm for conversational UX design

Abstract

User interfaces that take human conversation as their interaction metaphor work fundamentally differently than those that employ spatial metaphors, such as a desktop or a page. While the fundamental concept in visual interface design is the layout, the fundamental concept in conversational interface design is the sequence. Each provides for the overall structure of the user experience. In the past, user-interface designers have borrowed elements from the various areas of physical design. From industrial design, they have borrowed concepts such as, buttons, levers, wheels, and more from the print industry, they have borrowed the page, typography, iconography, illustration, and photography and more. These concepts from the physical world are then adapted to persistent, visual representations on a computer screen. Of course, virtual buttons are different from physical buttons and displayed words are different from printed words, but they evoke familiar ways of interacting with the real world that are then repurposed for a computer–user interface. And graphical user interface design, web design and mobile design are mature disciplines with shared standards and communities of practitioners. However, the spatial interaction metaphors of these areas of visual design largely do not apply to the design of conversational user interfaces (Moore & Arar, 2019; Moore et al., 2020; Murad et al., 2021; Yankelovich et al., 1995). Human conversational interaction consists primarily of sequences of words and embodied actions, not of layouts of visual elements. Buttons or pages cannot be represented as a stream of words produced by different parties. Conversational interfaces are more akin to command-line interaction, which involves sequences of specialized commands. The interaction conventions of visual design, graphical, web and mobile, were invented as an alternative to language-based interfaces and are not applicable to the design of conversational user experience. Where then can UX designers find inspiration when creating conversational interfaces with their sequences of natural-language utterances? In part, they can turn to Natural Language Processing (NLP), which provides mature methods for recognizing, classifying, and generating natural-language input and output (Chowdhury, 2003; Goldberg, 2017; Graves et al., 2013; López-Cózar et al., 2011; McTear et al., 2016; Reiter & Dale, 1997). These methods help designers understand what the user said and render realistic voice responses. However, NLP provides resources primarily for managing natural language, not for managing natural conversation. NLP addresses language use in any form: novels, poems, tweets, e-mails, conversations, etc. (Berg, 2015; Mitri, 2022; Peng et al., 2018; Zhang & Gao, 2017). Any bit of natural language, be it Spanish, English, Mandarin, etc., is analyzable with NLP.