Emotion recognition is a core element in the development of emotional intelligence in machines. Early models are designed to recognise full-blown emotions which are easily identifiable. However, we rarely experience these types of emotion in our daily lives. Identifying our own and others' emotions with confidence is often difficult: this is called emotional ambiguity. Databases are the building blocks of the development of emotion recognition systems, hence they should introduce ambiguity in emotional representation. This paper summarises the main emotional representations and proposes a literature review of the most used multimodal databases in emotion recognition, with a study of their position on the problem. The paper further discusses the possibility of representing emotion ambiguity from the selected databases.