- From: Thomas Steiner <tomac@google.com>
- Date: Mon, 4 Oct 2010 16:59:51 +0200
- To: tmichel@w3.org
- Cc: "public-media-annotation@w3.org" <public-media-annotation@w3.org>
Hi Thierry, hi Work Group members, Thank you for your detailed response! Please find my comments below. > 1) Subtitles > > Concerning external subtitles, using ma:relation is the correct approach as > in your example. The identifier attribute contains the URL of the subtitle > file, and the relation type qualifies it as a subtitle relation. The value > should be a URI, but could also be a string. It is recommended to use a > controlled vocabulary for the type of the relation. This is perfectly what I was hoping for. Agreed. > Embedding of subtitles is not a use case that we considered, however it is > possible. The mechanism we use to specify timed metadata is to specify > fragments identified by Media Fragment URIs [1] and then describe > annotations of these fragments. Same here. Agreed. > - Link to external subtitle file using ma:fragment, with type subtitle and a > Timed Text Markup Language (TTML) [2] or WebSRT [3] file as target. Assuming ma:fragment is actually ma:relation. Agreed. > - Subtitles can be embedded in a media file, in which case they can be > described as a track media fragment using ma:fragment and Media Fragment > URIs [1]. This sounds like a nice and flexible way. Agreed. > - Subtitles could be embedded by using ma:title with a type qualifier for > subtitle. A list of time media fragments is defined and each fragment is > annotated using ma:title. This, while possible, sounds like an overcharge of what ma:title was designed for. Personally I'd not go this way. > Although the last option is a way of embedding subtitles that is not a use > case we considered. We expect that in most cases a dedicated format such as > TTML or WebSRT will be used for the subtitles and referenced. Agreed. TTML and WebSRT are both very good standards. > 2) Semantic annotation > > As described above, time based annotations are a possible. Currently, two > cases are covered by the spec: > > - use ma:description for a textual description of the media resource (or a > fragment) Possible, but not machine-readable/understandable by default. > - use ma:relation to link to a RDF file or named graph containing the > annotation for the media resource (or fragment) Good solution. Especially when used with ma:fragment this has the potential and degree of freedom of expression that I need. Agreed. > There is currently no solution for embedding a set of triples into one of > the ma:* properties. We understand that might be useful and have started > discussion with the Semantic Web Coordination Group about a solution for > this problem (see thread starting at [3]). The summary of the discussion is: > Named graphs could be a solution to this issue, but there is no standard > syntax for expressing them, to which our specification could refer. Such a > syntax might find its way into RDF 2.0. As no other applicable solution > emerged in the discussion, we decided to exclude the embedding of triples > into ma:* elements until a standard syntax for named graphs is available. Let's continue the discussion of [3] (the 2nd [3] ;-) below the first [3]) at [3]. Cheers, Tom -- Thomas Steiner, Research Scientist, Google Inc. http://blog.tomayac.com, http://twitter.com/tomayac
Received on Monday, 4 October 2010 15:00:45 UTC