Human language is mainly carried by speaking and writing. Obviously, linguistics has made great contributions to artificial intelligence in teaching computers to recognize what people say (speech recognition), to teach computer-generated words (speech synthesis), to teach computers to understand what people write (natural language processing) and to teach computers to generate texts (natural language generation). To achieve the above four tasks, simply and rudely speaking, it is necessary to use statistical models to describe what human beings say and write. How should statistical models be obtained? It is trained by data. What are these data? That's a linguistic mark of words or sounds. Give a small example of speech synthesis. For example, if you want to synthesize a sentence "Dr.SmithlivesinNYC", the input is this text and the output is a waveform. The general idea is that you should find the phonemes that match the phonetic symbols through the linguistic annotation of the text and put them together.