Natural Language Processing
- need linguists to engineer
- to development new theory of NLP
- Machine learning (human generate the rule of rules)
- large amount of
- need automatic learning algorithms and big data
- not necessarily readable
- to refine the existing method
- deep learning
- In 1950, Turing came up with the idea of NLP.
- In 1960s, people started to use dictionary-way to translate languages to one another, which is very awkward.
- In 1970s, people started to create chatterbots.
- In 1980s, mostly people made complex sets of hand-written rules to analyze natural language.
- In 1980s, machine learning was introduced due to the increased computing power. Statistics model became popular in machine learning algorithms. The dominance of linguistics theory was lessened.
- During these time, supervised learning algorithms were most focused; such algorithms needs people to modify huge amount of data for it to learn.
- Nowadays, unsupervised learning algorithms starts to develop, though it is not as accurate and effective as unsupervised ones, it is the future of machine learning.
- Words tent to have different meanings in different context, an algorithm should be able to differentiate them.
- Sentence Structure
- Natural language has various forms that are not very strict, an algorithm should be able to recognize them.
- Traditionally, people tended to break a sentence into parts, and explain each part separately, then explain their relationship with each other.
- The computational way to recognize sentence culture is mainly through statical learning. People first a collection of sentences which have been parsed by human, and algorithms will learn the pattern from these sentences.
- People make mistakes, algorithms should be able to tolerate them.
- between Saying and Acting
- When people asked can you give me an example of something, it is awkward to answer yes, I can, rather the example should be given.
- Chinese's Unique problem: Recognize word
- In Chinese, there is no space between characters, so the algorithms have to figure out which several characters form a word.
- What is the future of NLP?
- Why it is hard to do unsupervised learning?
- How does deep learning applied in the NLP?