主要观点总结
文章介绍了Geoffrey Hinton对语言与认知的三个不同观点，并解释了符号、向量和嵌入在认知过程中的作用及它们之间的关系。文章还提到了机器翻译中的循环神经网络（RNN）和隐藏状态，以及它们在语言翻译中的应用。
关键观点总结
关键观点1: Geoffrey Hinton对语言与认知的三个观点
文章介绍了符号观、向量观和嵌入观，这三种观点代表了语言与认知的不同理解方式。
关键观点2: 符号、向量和嵌入的解释及关系
符号是语言的离散表示，向量是符号的数学编码，而嵌入是捕获符号之间语义关系的特殊向量。三者之间有着紧密的关系，它们在认知过程中起着不同的作用。
关键观点3: RNN和隐藏状态在机器翻译中的应用
RNN通过处理输入句子中的每个词并更新隐藏状态来捕获句子的意义。隐藏状态是一个向量，它捕获了句子的上下文和意义。这个向量然后被用来生成翻译句子。
文章预览
"Whether language evolved with the brain or the brain evolved with language?" is a question answered by Geoffrey Hinton, which I felt might be interesting to share and elaborate some explanation. 3 different views of language and how they relate to cognition 1st, symbolic view: cognition consists of having strings of symbols in some clean logical language where there is no ambiguity and applying rules of inference, cognition is just these symbolic manipulations on things that are like the language symbols. 2nd view. Once you get inside the head, it's all vectors, symbols come in, you convert these symbols into big vectors, and if you want output you produce symbols again. There is a point in machine translation in 2014, where people used recurrent neural nets(RNN) and words keep coming in and having a hidden state that keep accumulating information in the hidden state. So when they get to the end of the sentence, they have a big hidden vector that captures the meaning of that sentence
………………………………
