Building blocks of AI systems
In modern machine translation, this component compresses the meaning of a sentence into a vector.
What is an encoder?
This mechanism lets the decoder “look back” at specific input words while generating output.
What is attention?
Mitchell describes how this widely used translation system shifted from phrase based statistical methods to a neural network encoder-decoder model in 2016.
What is Google Translate?
For as long as this quality is lacking in caption generating, these translating systems will remain untrustworthy.
What is understanding?
This AI learning method relied on learning from data (such as large collections of sentences) rather than having humans specify rules in translation.
What is Statistical Machine Learning?
a program used in virtually all evaluations of machine translations that compares a machines translation with those of humans and spits out out a score
What is Bilingual Evaluation Understudy (BLEU)?
a problem that occurs when an import to an encoder network becomes too long.
What is the system "forgets"?
complex units proposed by a Switzerland research group that assigns weights to individual units in a recurrent neural network that allows for the memory last throughout the sentence
What is Long short-term memory (LSTM)?
System developed by Google that translates images into language
What is Show and Tell?
Two types of language when it comes to translating words and sentences
What is Source Language and Target Language?
newer translation technology adopted by Google in 2016 that was “bridging the gap between human and machine translation” using deep learning
What is Neural Machine Translation?
Online demo system developed by Microsoft that automatically identifies the objects in a picture, interprets what is going on, and writes an accurate caption explaining it
What is CaptionBot?