meta data for this page
This is an old revision of the document!
Understanding Speech Patterns
When using a grammar to interpret what a caller says in response to a prompt, it's important to take into account likely speech patterns.
For example, a prompt like, “Would you like to continue?” is an example of a yes/no prompt, but reasonable responses to it may well include words or phrases other than “yes” and “no”. Most “built-in” yes/no (aka, Boolean) grammars include alternatives to “yes” and “no” such as “yeah” and “nope”, but do not include more likely conversational utterances such as “Yes, I would” or “No, I wouldn't”. Note that the wording of the prompt may drive certain characteristics of the responses. If the prompt were, “Do you want to continue”, then it's unlikely that callers would say “Yes, I would”, but they might well say “Yes, I do” or “No, I don't”. Callers might also include politeness markers such as “Yes, please” or “No thanks”. (For more information about using yes/no prompts, see Two-Item Menus and Yes-No Prompts). Similar considerations apply when collecting other types of information. For example, if you're designing a reservation system and have asked for the start date, the grammar shouldn't require the caller to say the year, but should capture it if spoken.
Built-in grammars can be used where appropriate and with caution. It will often be necessary to either use the built-in grammar as a starting point but modify it as needed, or to just build the needed grammar from scratch after giving careful thought to the different ways callers might respond to the target prompt.
Similar considerations apply when collecting other types of information. For example, if you're designing a reservation system and have asked for the start date, the grammar shouldn't require the caller to say the year, but should capture it if spoken. When asking for information such as social security numbers and credit card numbers, the caller may preface their response with “my social security number is …” or “my credit card number is …”. Typically, such introductory phrases are not supported by “builtin” grammars but adding them as introductions may help to improve recognition.
When designing grammars to collect items such as company names, university names or country names, be aware that callers may either shorten (casual/colloquial references) or expand the name (formal/legal references) in their responses. For example, a caller may omit “corporation”, “incorporated” or “company” in a company name. For country names, the caller might use a formal reference such as “the Federal Republic of Germany” rather than just “Germany.” Adding shortened and expanded forms of names judiciously can increase the likelihood that the caller is understood correctly on the first try.
As an overall guideline, keep grammars simple then use tuning data to expand.