Wednesday, August 19, 2015
11:45 AM - 12:30 PM
|Level: ||Technical - Intermediate|
The usual focus when developing Virtual Assistants and analyzing their dialogs with human users, is on what the VAs say and how to make their part of the conversation as human-like as possible. In this session we want to bring more light into the human side of the conversation and focus on learning as much from it as possible.
We will present how we:
- Designed and implemented a flexible and extensible query language, specialized for real-time analysis of NLI dialogs
- Mine large numbers of VA dialogs for information about the participants, their profiles, opinions, motives and behavior
- Use machine-learning to integrate both automatic improvements and semi-automatic improvement recommendations into the VA development cycle
- Combine rule-based and statistical methods to enable free-form, natural language querying of natural language data
We will approach these topics from a pragmatic angle and share our experience on tackling the specific challenges of dialog NL data. The key points will be relevant and valuable independently of the concrete NLI platform that we have developed. The focus will lie on the iterative nature of the NLI application development and how the data produced by the application can feed the further application development.
Eric Aili is a multi-disciplinary, holistic R&D Engineer with a passion for computational linguistics, artificial intelligence and big-data.
At Artificial Solutions, provider of the Teneo Platform for development of natural language interaction applications, he has spent over a decade developing and researching different components of the platform for developing Virtual Assistants.
In the R&D team he works in close cooperation with language technology researchers to both invent and turn ideas and algorithms into robust, performant, scalable implementations.
He has been crucial in developing several patented and patent-pending techniques for automatic code generation and tuning for natural language interaction and for natural language data analytics.
Sonja Petrovic Lundberg is an R&D Engineer at Artificial Solutions, provider of the Teneo Platform for development of natural language interaction applications. With a team of co-workers, she has developed patented techniques for automatic NLI rule generation and tuning and for the operation of a virtual assistant network.
Ms. Lundberg has background in mathematics and organizational sciences from the University of Belgrade, Serbia, and has for the last 15 years worked on developing new methods in various fields of computational linguistics. She has studied language technology at Stockholm and Uppsala universities as well as the Royal Institute of Technology in Sweden.
Prior to joining Artificial Solutions, Ms. Lundberg led data-driven and classroom-independent computer-assisted language learning with the award-winning "lernu!" web-platform. On behalf of Esperantic Studies Foundation, she has worked on combining statistical and rule-based approaches for NLP and machine translation.