- From: Scott McGlashan <scott.mcglashan@pipebeach.com>
- Date: Thu, 5 Sep 2002 14:00:02 +0200
- To: "Guillaume Berche" <guillaume.berche@eloquant.com>, <www-voice@w3.org>
Guillaume, thank you for your input. Unfortunately, input on this issue was required within the official comment period which ended on the 24th May 2002 as is clearly stated in the specification. The group has already discussed this issue and it is too late to take your input into account. You, of course, will another opportunity to comment on the specification when the next version is published. Scott VoiceXML dialog team leader, VBWG -----Original Message----- From: Guillaume Berche [mailto:guillaume.berche@eloquant.com] Sent: Tuesday, September 03, 2002 16:04 To: www-voice@w3.org Subject: Comment on alternate prompt issue from section "4.1.3 Audio Prompting" Hello, This comment relates to the issue in section "4.1.3 Audio Prompting" of the VXML W3C Working Draft 24 April 2002: "if events can be thrown as the result of the prompt playing possibly unrelated to the input collection, this forces VXML authors to defensively code against possible audio failures unrelated to their content. [...]" I tried to clarify the need for such defensive code by listing such "use-cases": a - an existing application is executed onto a new interpreter which has different capabilities (such as different supported audio format, different supported language). The developer is trying to program its application to programmatically recover from such condition by using different format/language. b- in an application composed of multiple documents developped independently (such as a voice portal), a caller document has queued prompts which fail to be played while the called document is running. I believe that the VXML language should not try to support recovery of conditions such as case a), rather events should be used to help authors diagnose errors; the application itself should not try to recover programmatically from it. This is because VXML events as they are currently defined do not provide sufficient information to let an application programmatically recover from the reported error condition. I believe it is not desirable to try to support VXML applications that can dynamically recover from missing capabilities of their host interpreters and adapt from them using client-side logic. This would be too complex and probably cause performance problems anyway. In my opinion, two approaches can be followed to account for the different capabilities of the host interpreter: 1- Client-side approach: expose to the VXML document standard session variables that describe the capabilities of the platform. This way the application may degrade itself smoothly to the available capability of the interpreter. This is self-degradation which is different from programmatic recovery of failed prompt. 2- Server-side approach: follow the HTML model in which the client-side (the VXML browser) provides information about its capabilities, and the server-side dynamically adapts content to the "rendering" capabilities of the client. Therefore, a later version of the VXML specifications may precise semantics of standard HTTP headers (such as "Accept" which specifies the supported media formats or "Accept_Language") that specify the capabilities of the VXML host platform (such as supported language, supported audio formats...) Concerning the case b) described above, I believe there should be a way for application developers to enter a mode in which failed prompts are silently ignored. This could be achieved through an additional property (such as "ignorebadprompts"). This property would be resolved in the property scope of the form item which triggers entering of the waiting phase. Therefore case a) would simply be handled by defining the "ignorebadprompts" in the usual entry point of the document. It would be usually the case that the rest of the application does not define this property to preserve the automated detection of failed prompts from the application itself (which can trigger a submit element targetting a page sending an email report for instance). For the VXML specification 2.0, I would suggest to clarify the intent of the events thrown by the interpreter to precise they support application development and tuning, but the specs do not expect applications to recover from such conditions. Possibly document the future intent of client-side and server-side capabilities discovery as described above. Suggested text addition to section "4.1.3 Audio Prompting": "When a prompt fails to be played (for various reasons such as an unsupported language, or an audio prompt which can not be fetched and without alternative prompt), then the appropriate event is thrown. Note that the VXML language is not designed to support automatic recovery from the failed prompt. Rather, thrown event is designed to allow VXML authors to automate detection of errors and possibly their reporting (by catching the error and transitionning to a submit form item connected to a page which triggers an email notification on the server-side and returns an error page to the end-user). A possible extension of VXML in a future version would be to support client-side and and server-side browser capabilities discovery which lets both the VXML application (or application server generating VXML content) to dynamically degrade/adapt the application to the capabilities of the host interpreter. Note that the support for programmatic error recovery from failed prompts will probably not be enhanced in future versions." Please let me know if wording suggestion concerning the "ignorebadprompts" property would be useful. Any comment on this is welcome. Guillaume.
Received on Thursday, 5 September 2002 08:01:32 UTC