A discourse is series of utterances grouped together in either a monologue or a dialogue. Participants of a monologue are fixed in their roles of speaker and hearer, whilst the participants of a dialogue take turns at being the speaker or the hearer.
A dialogue normally consists of many different acts of communication such as asking a question, giving an answer, making corrections and so on, whereas there is no feedback from the hearer in a monologue.
Early dialogue agents such as ELIZA [29] forgo any complex form of NLP and relied on simple pattern matching and replacing of words. A true dialogue is very hard to maintain due to the world knowledge needed in order to understand the other participant of a dialogue.
Another early dialogue agent that was much more advanced than ELIZA was SHRDLU. This dialogue agent had control of a robotic arm and an eye on to the now legendary BLOCKS world [30]. This agent could carry out commands, and answer questions within its limited domain; a small platform containing coloured blocks and pyramids.
Problems such as creating a all encompassing dialogue agent have come to be informally known as AI-complete, one that essentially requires all the knowledge (and abilities to use it) that humans have [13].
Many problems occur in both dialogues and monologues which make discourses hard to understand. Most notably those of (co)reference and inference. In Section 2.2.3.3 modus ponens was discussed along with its use in both forward and backward chaining but there is also a third, invalid, use of modus ponens known as abduction, which is often used in the pragmatics domain.