Conversational AI 12-DIALOGUE COMPONENTS IN THE STATISTICAL DATA-DRIVEN APPROACH
NATURAL LANGUAGE UNDERSTANDING
The NLU module takes the output from the ASR module and produces a representation of the meaning of the user’s utterance. In spoken dialogue systems the NLU module is often called Spoken Language Understanding (SLU) and the term Conversational Language Understanding has also been used to emphasize that the nature of the language involved differs from that of more formal written text.
The main task of DM is to decide what action to take next given the user’s input and the current state of the dialogue. Traditionally, dialogue management has involved using rules handcrafted by a dialogue designer to implement predetermined design decisions. For example, handling potential ASR misrecognitions might involve considerations as to whether and when to confirm the user’s input and whether to use information such as ASR confidence scores. These design decisions, which are generally based on experience and best practice guidelines, are applied in an iterative process of design and testing until the optimal system is produced.
NATURAL LANGUAGE GENERATION
NLG takes the output from DM and converts it into text. NLG is important since the quality of the system’s output can affect the user’s perceptions of the usability of the overall system. Corpus-based methods have been used to optimize the output of NLG. Using a corpus of suitable data, such as the utterances of a domain expert, has the advantage that the generated text will be of good quality. In this approach, known as over-generation and re-ranking, a number of candidate outputs are generated and the best one is selected based on a re-ranking algorithm.