| Popis: |
By definition, the goal of NL speech research has been “natural” human language—that is, faithfully simulating human-human conversation with human-computer dialogues. The word “natural” here is enclosed in quotation marks to emphasize its narrow definition. Attempts to explore alternatives to this narrow definition have attracted little interest. But there are other ways to incorporate speech into effective and intelligent user interfaces. Text-to-speech synthesis, for example, can easily exceed human capabilities—encompassing a wider pitch range, speaking faster or slower, and/or pronouncing tongue-twisters. Similarly, non-speech audio can provide prompts and feedback more quickly than speech, and can also exploit musical syntax and semantics. Speech recognition dialogues, in turn, can master pidgin languages, shortcut expressions, cockney-like underground tongues, and non-speech sounds. User interfaces designed around these super-human technology capabilities are known collectively as super-natural language dialogues (SNLD). Such interfaces are inherently multimodal, and their study is inherently interdisciplinary. |
