PANGEANIC’s proposal to advance the current state-of-the-art and incorporate new processes to artificial neural networks for service in industry has been awarded a development grant by the Center for Industrial Technological Development (CDTI) as “NEURONAL AUTOMATIC TRANSLATION PLATFORM”. CDTI and the European Union will provide financial support through the Intelligent Operational Growth Program and in collaboration with a world-leading research center for pattern recognition, the PRHLT of the Polytechnic University of Valencia.
Pangeanic’s hybrid neuronal machine translation platform will offer advanced software based on neuronal networks for automatic translation processes through the development of new hybridization techniques. This use involves the development and application of techniques from artificial intelligence that will provide added value to both professional users and end clients needing to process Big Data in other languages, machine translate billions of words of online content, quickly and efficiently and, increasingly, publish it with little human post-editing effort.
- Development of hybridization techniques between neural network cores and statistical translator Moses (SMT or MMT).
- Development of APIs capable of using different neural or hybrid translation cores, as well as adding new cores in the future.
- Development of management of APIs to allow the intelligent selection of different pre-processes, post-processes and translation scripts with a high degree of autonomy according to needs.
- Development of optimized training and retraining processes for neural translation engines, maintaining a specific set or intelligent hybridization options of neural translation and statistical translation technologies .
- Automating customization in machine translation engines according to a particular user own style and terminology needs (adaptive training) in separate instances.
- Development of models that allow the creation of translation engines without the need for extensive knowledge of natural language processing.
- Metrics-based optimization of translation engines.
APPROVED BUDGET: 348,705 € – PROJECT NUMBER: IDI-20170964
FINANCING GRANTED: PARTIALLY REIMBURSED AID OF 85% OF THE BUDGET.
What is an Artificial Neural Network?
An Artificial Neural Network (ANN) is a paradigm that processes information in a way that can remind us of our own biological nervous systems (with interconnected decision centers taking place and weighing a decision). The key element of this paradigm (let’s remember that ANN are not new) is the structure of the information processing system, which in the case of human beings is composed of a large number of highly interconnected processing elements we call neurons.
Not only human beings are capable of making a decision. Many mammals can also, based on intuition, experience and “something” that we typically don’t understand, but which we call a kind of intelligence. Facing a problem, neurons work together to solve it, each adding a piece of information. Artificial Neural Networks learn through examples, just like we do – after all, that is what we can experience. And to gain this experience, neural networks need samples, that is, data, massive amounts of data. A neural network can be configured to learn many things, not just bilingual patterns between two languages (two systems). It can also be used for any type of pattern recognition such as data classification (images, objects, shapes, etc.), handwriting or providing set answers to emails, or a chatbox.
Basically, a neural network requires a learning process, after which it can create quite precise representations of the queries. Taking the human (or mammal) example, a biological system requires adjustments to the synaptic connections between neurons – and this also happens in artificial networks.