In this paper, we proposed a two-phase project on emotion corpus creation based on multi-knowledge of cognitive semantics, discourse analysis, paralinguistics, and computer science. Data were gathered from Thai lexicon of five main Thai dictionaries and thesaurus, in addition to written and spoken texts of people with depression in Thai and facial expression with speech situation. We found that semantic primes and features of each emotion were needed to serve as a guideline of emotion categorization in Thai context. We introduced the step-by-step methods of the first phase to create Thai emotion corpus entailing both verbal and nonverbal corpora. The way to classify emotion corpus by focusing on the specific text of depression as well as to find the guidelines of labelling facial expression in the situation of specific emotions was explored. Lastly, the step of creating emotion corpus in the second phase was introduced with some suggestions and discussion.