Ai Dreams Forum
Artificial Intelligence => General AI Discussion => Topic started by: infurl on April 22, 2022, 02:03:14 am
-
https://www.amazon.science/blog/amazon-releases-51-language-dataset-for-language-understanding (https://www.amazon.science/blog/amazon-releases-51-language-dataset-for-language-understanding)
Imagine that all people around the world could use voice AI systems such as Alexa in their native tongues. One promising approach to realizing this vision is massively multilingual natural-language understanding (MMNLU), a paradigm in which a single machine learning model can parse and understand inputs from many typologically diverse languages. By learning a shared data representation that spans languages, the model can transfer knowledge from languages with abundant training data to those in which training data is scarce.
A long time ago I decided to focus my efforts on natural language understanding. It is only one aspect of intelligence but I figured that our natural languages have evolved to model what we think about. Therefore, if I could write software that could model what we think about, I would have solved a large part of the problem.
-
did you hear about this?
https://ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html
-
NLU, NLP, speech recognition is all still undiscovered science. Since the Hidden Markov Model (statistical analysis) began in the 1980's, true scientific study stopped, and was replaced by giving the machine "enough information" to then guess from...
Still undiscovered:- What links words together in the most broad sense of synonyms, such that changing language mid sentence is possible and replacing blank words with "something"/"someone" (etc).
- What links sentences together in the most broad sense to cover any language at once, with similar sentences, to output broad intentions.
- Reverse looking-up the above in finding missing words/homophones in speech recognition/speech-to-text.
- How can you track real sound waves in real time without guessing, with valid ranges, discovering accents, replacing accent-based vowel substitution with common types. Etc.
Throwing more power at an unsolved problem doesn't help smart watches/glasses/vr & ar devices.