New Linguistic Intelligence Platform for Deep Reinforcement Learning
Deep Linguistic Parsing as a Game
JERUSALEM, February 15, 2018 (Newswire.com) - Linguistic Agents Ltd., a leader in Natural Language Processing, announced today a new and innovative platform for the training of deep reinforcement learning agents.
Clear as Mud
Because of recent stories of the spectacular success of deep reinforcement learning in various fields such as self-driving vehicles and the game of chess, many researchers are exploring ways to apply it to their respective areas of interest, and the field of natural language is no exception. However, the objective of defining natural language in terms of actions, states, and rewards is far from obvious.
This is why, after many years of concentrated effort directed at expressing abstract scientific theories as computational algorithms for use in the digital world, Linguistic Agents is in a unique position to address this challenge.
Accumulation of Insight
Not everyone in the computer science community believes that theoretical linguists working within the Chomskyan paradigm have come up with enough fully satisfying answers about the true workings of human language, but even if that is the case, they have at least come up with many important questions. Paradoxically, from the perspective of reinforcement learning, the accumulated insightful questions are even better than answers in that they are arguably more helpful in designing rewards for agents.
In order to use our proposed platform, reinforcement learning agents should be initialized with standardized language-oriented features before starting the process of training in their respective areas of specialization, After achieving proficiency in their particular field of expertise linguistically initiated robots can then return for learning additional natural language skills. In other words, they may from time to time upgrade their linguistic capabilities by both (a) receiving upgraded language models, and (b) training with an updated version of a robot “teacher”.
This new platform addresses a real problem: AI developers working on specialized applications would like to continuously maintain the up-to-date language models, but they're too busy! They need to concentrate on their projects and do not have time for this huge amount of work that would be required of them in order to stay in sync with the constantly evolving state-of-the-art in language technology.
The Linguistic Agents platform aims to shield AI developers from the unnecessary burden of having to become linguistics experts themselves.
Source: Linguistic Agents Ltd