• Context

Automatic Speech Recognition is a module that transcribes speech to text. Its robustness, accurateness, and universality have greatly improved these last years, so that it is now possible to use it with a generic language model, even in a noisy environment. This technologic advance pushes numerous innovations, including the one tackled by this PhD thesis: active listening to a human-human conversation to extract meaning and deliver a contextual service related to this meaning.

Internally to Orange Labs, we have already worked a service of this kind in the contextualisation service that is integrated in the work environment of call user agents. This environment embeds several professional services: chat, e-mail, call management, appointment scheduling, location, Customer Relationship Management, scripting, statistics, etc. The contextualisation tool consists in guiding the agent by launching semi-automatically or fully-automatically applications according to conversational patterns detected during communication.

As well, a personal assistant named MindMeld (http://www.expectlabs.com/mindmeld/) has been commercialised. It listens to phone calls and proposes in real time contents related to approached subjects.

The PhD thesis subject will focus on the dialogical part, namely from one side, not only recognize the keywords or even analyze sentences, but placing analysis in a human-human dialogue model ; and from the other side, not only display an information or propose an action, but start a dialogue with the user. Both fields have never been investigated at this time to our knowledge.

The scientific domain of the PhD thesis will be split into semantic analysis of human-human dialogue and human-machine dialogue systems. The latter include interactive voice response systems, but also chatbots or interactions that some video games propose between the player and some characters. Spoken dialogue systems are expanding and deal commercially several billions calls a year.

• NADIA team

We are a small unit of 8 permanents, 3 PhD students, and a post doctorate researcher. The team is in charge of Orange dialogue technology: Disserto suite. Disserto has been used to develop dozens of interactive voice response systems, which receive more than 100M calls a year. Some Disserto-based multimodal prototypes have been released in the research side as well. In addition to its strong industrial implication, NADIA is also involved in academic research with 3 conference articles published in 2013 and already 4 accepted in 2014.

• Scientific objectives

PhD thesis scientific thematic revolves around three main disjointed subjects:

• How to model semantic analysis of a human-human conversation and extract a meaning further used in an application?

• How to initiate, lead and conclude a dialogue with a user that is already committed in a conversation with someone else?

• How to make this kind of listening and dialoguing application easily extensible, and even customizable?

We will ask to the PhD student to endorse the prototype development of several demonstrators and to organise experimentation campaigns to evaluate the results of her/his work.

On the other hand, the following subjects are beyond the scope of the PhD: automatic speech recognition (we will use a plug and play product) and multimodality theoretic studies. Both are very large fields of research that must be studied separately.

• specialities,

• Spoken Dialogue Systems

• Human-human dialogue theory

• Natural Language Processing

• Machine Learning

• Required educational level,

• Research master of science or « école d’ingénieur »

• At least one of the cited specialities recorded inside the university curriculum

• Required experience,

At least an internship in one of the cited specialities, or more largely in Artificial Intelligence

• Attractiveness of the position

This PhD thesis offers the opportunity to contribute to the future of dialogue applications, and to participate to putting them to everyone’s reach in everyday life. Those two domains are currently fast-growing and their emergence seems ineluctable in the ten years to come. According to Gartner, personal assistants are among the top10 technological trends for 2014, inside the “smart machines” thematic.

The PhD student is integrated in a small team comprising three other PhD students and a post-doctorate researcher. All team works in the dialogue system field. The other Orange Labs teams gather every kind of artificial intelligence skills and more largely in computer science. It is the perfect environment to lead significant research and get quick and high level advice from experts.

Orange warrants that the PhD student will work in the best conditions by making several commitments:

• Provision of a state-of-the-art industrial software: our historical tool, namely Disserto, will be used by the PhD student. It will enable her/him to tackle the PhD issues without spending to much effort on peripheral development.

• Provision of our previous research results: research and market watch, theoretical studies and code will be made available. Orange Labs gather 3700 researchers. Whatever the research study is, it is likely that somebody has prepared the ground.

• Provision of necessary equipment: to study dialogue in home environment, we have already invested on equipment to build a showroom.

• Experimentation programs: we are used at Orange Labs to launch user test campaigns on our internal development. Each campaign is specific, but to give an idea, my first PhD student launched three experimentations with 500 unique testers (and almost 2000 dialogues)

Provision of data: Orange has a big experience in spoken dialogue systems and has a large amount of data available (Disserto-based systems receive more than 150M calls a year).