Since the introduction of Deep Learning, machines got the powerful ability to learn something. Natural language processing (NLP), which is the language analysis system, is also getting improved thanks to the deep learning method. This is the first article in the series of “Machine That Understands our Languages.” In the series, we will explore how a machine understands our natural languages.
Why Are Our Languages Hard to Understand?
Before deep learning was developed, for a computer, it was so hard to understand our languages. I have learned English as my second language, and I studied English vocabulary and grammar. While I was in English grammar class, my teacher told us something. According to my English teacher, English language is so illogical and this is kind of crazy language. Why do we say “I am, you are” instead of “I is, you is”? But, a computer thinks logically.
Deep Learning and NLP
A deep learning network is a type of neural networks and constructed by many artificial neurons (which is influenced by biological neurons or nervous cells). Usually, the network is built by writing a programming code. This is kind of normal function that has inputs and outputs for programmers. However, the reason this becomes the powerful tool is because of its learning ability. By giving data-sets that have input data and the targeted output data, Deep learning networks learn as we, humans, obtain experiences. Finally, the network can predict the output for new data it has never experienced yet.
NLP also had developed rapidly thanks to the deep learning. There is a common language analysis, which is the part-of-speech tagging. This is a labeling process of each word depend on the role of the word. In other words, the analysis system detects whether or not the word is a noun, verb, adjective, etc. A few decades ago, programmers had to program all cases for all words in order to label. But, providing some sample data to deep learning network, it became possible to detect the part-of-speech with high accuracy and less effort.
What Does It Mean By Understanding?
First of all, the neural network that constructs deep learning network was created by mimicking our brain. Thus, we need to think about what exactly understanding means. Understanding is our ability we usually use when we study, watch movies, talk, etc. How do we understand our languages?
Definition 1: Understanding languageEvery time we understand something, we react in some ways. For example, when your friend talks to you about US President Donald Trump, probably you show some kind of respect for him, or disrespect. This is the response (output) of the conversation (input). Obviously, you can response because you know who Donald Trump is and what you and your friend are talking about. In short, you response after you understand. Even if you don’t react directly, your body may show some signals such as nodding.
The ability to understand language is the ability to response or decide what to do next. “What to do” includes any motion, imagination, generating thoughts, memorizing, etc.
Watching a movie is another kind of understanding. We usually don’t response directly to the movie while watching. But, we still understand the story. Considering the definition 1, while we are watching a movie, we are memorizing the story. Probably, our brain remembers the story with some events. At a moment, an event happened in the movie. At the next moment, another event happened. Our brain’s job is connecting those events in order and storing some places in the brain.