Phoster

Research and Development

Crowdsourcing Dialogue Systems

Introduction

Some technical topics are broached toward the crowdsourcing of dialogue system content and behavior.

Collaborative Authoring

The content and behavior of a dialogue system can be represented in a number of ways.

Firstly, the content and behavior of a dialogue system can be represented in programming language source code files. Collaborative authoring, in this case, is a matter of integrated development environments and source code repositories, version control systems.

Secondly, the content and behavior of a dialogue system can be separated from source code files as data stored in some data format or in a database. Collaborative authoring, in this case, could require custom software tools.

Thirdly, a number of services, cognitive services, can encapsulate the content and behavior of a dialogue system. Collaborative authoring, in this case, could require utilization of such services or related user interfaces.

Fourthly, the content and behavior of a dialogue system can be represented as a set of interrelated, URL-addressable, editable pages. Servers can provide content for a number of different content types, for example hypertext for content authors and other formats for dialogue system user agents. Server-side scripting can be utilized to generate pages and generated pages can contain client-side scripting.

Fifthly, the content and behavior of a dialogue system can be represented as a set of interrelated, URL-addressable, editable diagrams.

Sixthly, the content and behavior of a dialogue system can be represented in transcript form. Transcript-based user interfaces may resemble instant messaging applications, scrollable sequences of speech bubbles, with speech bubbles coming from the left and right sides, such that users can edit the content in dialogue systems’ speech bubbles. Users could opt to view more than plain text in speech bubbles. There could also be vertical, colored bands in one or both margins, visually indicating discourse behaviors, moves, objectives or plans which span one or multiple utterances.

Collaborative Debugging

Debugging dialogue systems is an important topic. Debugging scenarios include switching from interactions with dialogue systems to authoring processes such that dialogue context data is preserved.

Natural Language Generation and Understanding

Natural language generation can produce editable structured documents from the data stored in databases and knowledgebases. Generated content can contain, beyond natural language, data and program logic to facilitate the processing of constrained or unconstrained edits. Edits to generated content can result in changes to stored data.

Computer-aided Writing

Computer-aided writing can convenience content authors and assure quality. Software can, generally speaking, provide users with information, warnings and errors with regard to tentative edits. Software can support users including with regard to their spelling, grammar, word selection, readability, text coherence and cohesion. Software can measure the neutral point of view of natural language. Software can also process tentative edits with regard to their logical consistency with respect to data stored in databases and knowledgebases.

Wiki Dialogue Systems

Exploration into the collaborative authoring and debugging of dialogue systems could result in new wiki technologies. Wiki dialogue systems could resemble spoken language dialogue systems with transcript-based user interfaces, users able to easily switch between dialogue-based interactions and the editing of dialogue system content and behavior.