My (and at least some other people's) initial impression of the mapping of FRBR group 1 entities to Bibframe was that it would be something like
Work = work + expression
Instance = manifestation
It appears from the actual examples, that the mapping is more like
Work = work
Instance = expression + manifestation
Holdings (annotation) sort of = item
Interestingly, this essentially two-level mapping is very similar to what OLAC did for our prototype interface for moving images (https://blazing-sunset-24.heroku.com/).
Movie = work + primary (usually original) expression
Version = current expression + manifestation
We had a table for libraries and items were modeled as a relationship between libraries and versions (manifestations), which I think is essentially similar to Bibframe's holdings. The attributes of the items could then be hung off the relationship.
The reasons we took this approach were practical. Most of the attributes of expressions for commercial videos are what I think of as independent variables. That is, the fact that this DVD has a French subtitle track has no necessary connection to the fact that it has a full screen expression or to what other language options are available. For every new manifestation, the individual values for these types of expression have to be verified anew and linking up to some sort of existing expression record would save no time over just adding them to the manifestation record. This two-level approach (we presented item location as a version attribute) also worked well for display to the public.
However, it turned out that there were a couple situations in which this model did not work so well.
One is when there are multiple works on a manifestation and the expression values (such as language) related to each work vary. There was no easy way in our model to represent this.
For example, the English and Spanish language version of Dracula from 1931 are often packaged together.
Work 1 Expression 1 Manifestation
Dracula (1931) English soundtrack DVD (1999)
English French subtitles 1 disc
Work 2 Expression 2 OCLC# 46829789
Dracula (1931) Spanish soundtrackSpanish English and French subtitles
Without a separate expression level, it is unclear how to prevent the wrong connections from being made (work 1 has English subtitles or work 2 has an English soundtrack)
Work 1 Version
Dracula (1931) DVD (1999)
English 1 disc
Work 2 OCLC# 46829789
Dracula (1931) English soundtrack
Spanish French subtitles
English and French subtitles
The second case is when the expression isn't really a single independent variable (or couple of closely related ones such as French Dolby surround soundtrack), but rather a cluster of attributes that are inherently related and need to be reused together. For commercial videos, these are usually distinct intellectual or artistic versions (rather than things like dubbed soundtracks that are meant to be substitutions for accessibility). For example, a director's cut would usually have a duration associated with it and we might also know of a date or an editor. It might also need its own summary and would be connected to its own reviews or other annotations.
Work Expression Manifestation
Blade runner (1982) Final cut (2007) DVD (2007)
117 min. 2 discs
Review: ISBN 9781419850028
http://goo.gl/UgMQe OCLC# 173522015
There are also rare cases where even for information that we would normally consider as an isolated, independent variable, there is additional information that one would want to keep together. For example, many of Miyazaki's animated films have been dubbed into English with big name voice casts. I once came across a Criterion Collection DVD of a Japanese film that offered for comparison two different English subtitle tracks translated by two different scholars.
Expressions that consist of a cluster of related attributes are particularly important for musical expressions (performers, conductor, location, date, arrangement) and also some literary works.
It is also unclear to me whether it is possible to realize the full potential of RDA without the ability to encode all the FRBR group 1 entities separately.
I can see why the focus on translation from MARC led to the existing model. It is clearly the most practical approach for legacy data. Although many researchers have tried, no one has found an effective way to automate the identification of expressions in legacy data. It is not always possible even with manual review.
However, it seems to me that Bibframe does need to support the separation of all the WEMI entities, as well as the best possible environment for entering new data going forward. Perhaps there could be some parallel way to allow the creation of a Bibframe work record for an expression with an instance record that only describes the manifestation and that is linked as follows:
Bibframe Work (FRBR work) --> Bibframe Work (FRBR expression) --> Bibframe Instance (FRBR manifestation)
I also wonder how hardcoded the mapping of attributes to Bibframe classes is going to be. For example, there was a post that suggested that actors would probably be mapped to instances. For film actors, this is counter to the approach that makes sense to the moving image cataloging community. The majority of film actors should be associated with the work. This also makes sense from the point of view of efficient data modeling since we want to reuse the list of actors from the work record in all instances rather than recording them redundantly at the instance level. Will there be any mechanism in Bibframe to accommodate differing viewpoints such as these?