Изучить OpenCog

[Nashev]

https://ru.wikipedia.org/wiki/OpenCog https://github.com/opencog https://github.com/opencog/link-grammar https://habrahabr.ru/post/160255/ - OpenCog — проект создания ИР (AGI), Ярослав Логинов

https://m.hh.ru/vacancy/23214057:

Responsibilities: Develop algorithms for unsupervised language learning for English, Russian, Chinese and other languages Implement algorithms as C/C++/Scheme/Python code for open-source project http://opencog.org/ Requirements: Be able to study existing code re-factor and maintain it Be able to learn new and domain-specific languages Have working knowledge for at least some of the following and be able to learn the rest: C/C++ Scheme Python N-Grams: https://en.wikipedia.org/wiki/Grams_(search) Spanning trees: https://en.wikipedia.org/wiki/Spanning_tree Link grammar: https://en.wikipedia.org/wiki/Link_grammar Word-sense disambiguation: https://en.wikipedia.org/wiki/Word-sense_disambiguation Word2Vec: https://en.wikipedia.org/wiki/Word2vec Neural networks: https://en.wikipedia.org/wiki/Artificial_neural_network Terms: Full time Partially may be telecommuting/remote Ability to read, write and understand spoken English Ph.D. study may be possible and welcome in scope of work Flexible schedule is possible Project is expected to be long term

[Nashev]

https://blog.singularitynet.io/singularitynet-integrates-opencog-s-atomspace-hypergraph-to-accelerate-intelligence-development-1301e736766d реализуют следующую идею для unsupervised language learning: опираясь на короткие фразы-словосочетания, смотреть какие фразы встречаются в предложениях, содержащих всего на одно слово больше чем ранее «усвоенные» предложения. Так, возможно, проще интерпретировать как эти новые слова, так и слова в исходных простых предложениях.

«To combat these challenges, we are starting with smaller lexicons with limited sentence lengths. First, using bi-term constructions of intransitive verbs, then with low-ambiguity three-term constructions with transitive verbs. We will keep increasing the richness of lexicon and length of the sentences, adding adjectives and adverbs.»