Once more about Wagner's operas

Tags: #<Tag:0x00007f7566d4d978> #<Tag:0x00007f7566d4d8b0>

The topic of how to split Wagner’s operas on parts was already discussed, for example here, but looking on the current state the operas there were no conclusion. At the moment a mix of various contradictory styles is used:

  • In some cases works are split on parts according to the composer’s intent, even if parts are relatively large. Example: “Die Walküre” is split on Acts and Scenes, which corresponds to parts indicated by the composer in the score.

  • In other cases works are split on very fine parts, where a new part is created nearly each time when a different performer sings. Example: “Tristan und Isolde”, Act I, where some parts are as short as 10 seconds.

  • Finally, sometimes works are split according to recordings of particular releases which does not necessary correspond to anything in the score. In worst case, the same Act may be split on sub-parts in parallel according to several releases. “Rienzi”, Act II looks like at least 3 different splitting schemas were applied simultaneously.

For most composers, the first splitting schema (parts according to the score) is undisputed, but in case of Wagner at least some editors seems to dislike it, probably because parts tends to be pretty large. In most releases recordings are smaller as such parts, and must be marked as “partial recording”.

The second schema (very fine-grained parts) allows to prevent partial recordings, but as a result most recordings will be linked to many very small parts. A typical recording is around 2 to 10 minutes in length, and with a very fine-grained parts a recording may be easily linked to 5 - 10 such works. Besides, even with relatively small parts the splitting is pretty arbitrary.

The last approach, which is even hard to name a “schema”, allows to link each recording to exactly one work “cut out” from the whole opera exactly for this particular recording. A disadvantage is that the number of parts explodes, many parts have intersections, and there is no clear logic behind such approach. In many cases, new works (parts) have to be created each time when a new release is added.

I prefer the first approach (parts according to the score, as indicated by the composer) to be used uniformly. This approach respects composer’s intent, and is not as arbitrary as the 2nd (fine-grained parts) or 3rd (follow recordings).

Before changing anything, I would like to hear your opinion. Does it make sense to uniformly apply one approach to all Wagner’s operas? If yes, which one?

If the decision will be to apply the first approach (parts according to the score as explicitly indicated by the composer), I volunteer to spend a couple of months cleaning things up.

To not be too generic, this question is solely about Wagner’s operas. It does not apply directly to any other composer, where there may be good arguments to apply another partitioning schema.

Edit: added last voting option to keep everything as is.

  • Parts according to score as explicitly indicated by the composer
  • Fine-grained parts
  • Parts follow recordings
  • Keep parts as they currently are, that is do not apply any uniform splitting schema

0 voters

No, Die Walküre is split into acts, scenes, and then parts of scenes. It has some of the same issues of fine-grained splits as you point out Tristan und Isolde as having. For example, see @Jim_DeLaHunt’s annotation on Act II, Scene II.

1 Like

This conversation is going to get extremely confusing very quickly, because it’s difficult to find wording that clearly expresses meaning. In particular:

  • Distinguish between what structure current Musicbrainz entries form, versus what current Musicbrainz style rules say should be the structure, versus what you propose as different style rules about the structure;
  • Distinguish between the word “part” meaning the Musicbrainz Part-of-work Relationship, and “part” meaning the usual English language meaning of a “portion of” or “component of”.
  • Use the active voice and be explicit about who or what does things in your descriptions. When the active voice is used, it is harder to tell who or what is “doing”.

So, when you say,

Consider saying instead,

…For some Musicbrainz Work entities such as work/c3869 “Die Walküre” , there are Part-of-Work relationships to Work entities which correspond to Acts and Scenes as specified by the composer in the score. This follows current CSG. But in other Work entities such as work/30a37 “Tristan und Isolde”, Act I, there are Part-of-Work relationships to Work entities which start on very fine-grained boundaries in the score, practically each time a different performer sings. This does not correspond to the CSG.
I would you like to poll your opionns on what structure the CSG should call for instead…

It is difficult and tedious to write this way. But without it, I fear we won’t be able to understand what points each of us is trying to make.

1 Like

So here is my opinion about the question of creating MB Work entities and relating them to MB Recording entities.

First, the contents of the MB database are the result of different MB editors at different times following the guidelines with different amounts of success, and sometimes even with different versions of the guidelines. We should expect that the database will forever have some entries that don’t follow the guidelines. The question in those cases is, always, how much work to expend to change the content to follow the guidelines. To a first approximation, the next editor to touch the content will probably improve it, but usually won’t get the content to follow the guidelines perfectly. And to a first approximation, it is always welcome for an editor who wants to, to clean up content so that it more closely follows the guidelines.

Second, I think the MB guidelines don’t promise that each Recording entity will have a single Work entity that corresponds to the music in that Recording, and only in that Recording. The guidelines define Work entities in terms of the composer’s and arranger’s activity. The Recording is a result of the performer’s choices, and the mastering engineer’s track divisions. They can differ. In classical music and opera especially, we should expect that they differ. The MB guidelines let us relate a Recording entity to a portion of a Work entity, or to multiple Work entities as a medley. We should use that capability.

See for example Release/da018 Die Walküre. The editor who made the Recording-to-Work relationships, @BitPerfectRichard, made them partial relationships, from each track’s Recording to a “partial” of the Work entity for the Act and Scene, not to any Work entity which is more fine-grained. So, Recording/49e01 Die Walküre: Aufzug II, Szene II. "Schlimm, furcht ich" relates to a “partial” of Work/f02f4d Die Walküre: Act II, Scene II, rather than to the fine-grained Work/ba47f Die Walküre: Act II, Scene II. "Schlimm, fürcht ich, schloß der Streit". I like this choice. I think it is a good way to link the content, given the present MB structures and guidelines.

Third, a virtue of the present MB guidelines for Work entities is that they let multiple editors independently arrive at the same results. Editors can agree on a single “correct” structure. Because the guidelines tell us to base Work entities on the music score, rather than on the boundaries of Recordings, then different decisions by mastering engineers about track divisions won’t confuse the choice of Work entity structure for a composition.

Fourth, I think the MB structures around compositions, Work entities, performances, and Recording entities are incomplete. There should be structures which correspond to the classical music practice of referring to a portion of a composition by the first important line of sung words. These structures should accommodate the practice of including introductory music and singing even before the first important line. The structures should accommodate the reality that different Releases will break performances into tracks and Recording entities according to local preference, so we can’t ever expect a single structure to fit all Releases.

Fifth, I think we should choose wisely about how much effort we should spend improving the MB structures, entities, and guidelines; and how much effort we spend improving content to fit existing MB structures; and how much effort we spend improving MB tools so that our existing content and structures are more useful for MB’s clients.

Sixth, I think our decisions about MB structures and guidelines should be driven by use cases and by value delivered to clients. We shouldn’t make changes in isolation. We should first identify who we will be helping, and what value we will be delivering to them. This should shape the changes we make. Not having this should make us cautious about changes.

My interest in modifying the guidelines is not great, I’d rather spend my efforts and attention span improving the worse parts of the existing content, and adding new content well. But show me clients who will get value from modified guidelines, and you might get me interested.


Not exactly clear what you are suggesting here. Something like “Aria from”? Examples? I’m looking at Puccini on MB, where the structure seems to be based around the arias.

“Aria from” sounds more like “excerpt” structure already supported by MB, when there is an “unofficial” (that is, not explicitly indicated by the composer in the score) part of a work which is often performed as a stand-alone piece. Example: Herman’s aria “Что наша жизнь? Игра!” from Tchaikovsky’s “Queen of Spades”. It is not a distinct part, but it’s very famous and is often recorded separately.

To my shame I am not familiar with Puccini’s works, but when comparing operas of Wagner with those of Tchaikovsky whom I know pretty good, there is a significant distinction in the structure. Tchaikovsky himself provided splitting of Acts on relatively small parts (“numbers”). In most cases those numbers have further natural split point on which everybody agree. For example, if Tchaikovsky himself named a number “Scene and Herman’s Arioso”, there is a further natural split on “Scene” and “Herman’s Arioso”. Since splitting on parts is very natural, and resulting parts have a typical length of 2 to 10 minutes, in most cases all releases follow the same structure when splitting on recordings. Differences appear only in seldom cases when a part indicated by Tchaikovsky is pretty large, and there is no natural division. In such seldom cases “partial recording” relationship solves all problems.

With Wagner the opposite is the case: most parts indicated by the composer himself are too large for a single recording, and usually there is no natural sub-division, so each release editor splits Scenes or even Acts on smaller parts as he sees fit. If the smallest part in MB is a “Scene” as defined by Wagner, most of recordings will be marked as “partial”. On the other hand, if the smallest part in MB is the lowest common denominator (“if any release split at this point, than we split at this point as well”), than most of recordings will be linked to several works. Whereas both cases (partial recording and linking to multiple works) are supported and perfectly OK, for most composers such cases are exceptional, but for Wagner either one or another will be the norm.

I suppose what @Jim_DeLaHunt means is an indication of all potential split points (assuming sane editors who do not split in a middle of a sentence), that is what I mentioned above as “if any release split at this point, than we split at this point as well”.

First of all, thanks to @Jim_DeLaHunt for a very detailed answer.

It looks like nobody takes seriously an approach “split by recording”, so we are down to two choices: coarse-grained parts indicated by the composer, or fine-grained parts defined by MB editors. There is also a possible kind of a hybrid approach mentioned as “Solution #3” below.

Solution #1: use as “MB work” only coarse-grained parts explicitly indicated by the composer in the score, merge all existing smaller MB works. In case of Wagner smallest parts indicated in the score are usually scenes, with typically 3 to 5 scenes in each act. Usually scenes are too large for a single recording, so we have to mark most recordings as “partial”.

Solution #2: split acts directly on fine-grained MB works, as it is currently done in Tristan und Isolde, Act I. Since recordings are usually larger, we have to link most recordings to several fine-grained MB works.

Solution #3: a mix of both. Split acts on coarse-grained MB works which corresponds to parts explicitly indicated by the composer in the score, but add one more level of fine-grained MB works underneath, as it is currently done in Walküre. I was wrong about the “Walküre” in my original posting when I thought it uses the solution #1. This solution may be considered a compromise: editors who prefer MB works identical to parts defined by the composer himself may use them, whereas editors who prefer fine-grained MB works may go a level deeper.

I prefer the first solution: keep only coarse-grained MB works and merge currently existing fine-grained MB works. Why?

  • Fine-grained MB works are either arbitrary (because divided only at some, but not all potential split points), or their number becomes insanely huge. Currently fine-grained works are mostly arbitrary, there is no clear guideline where to split, so two editors may easily disagree. An example is currently existing part 25 of Tristan und Isolde, Act I, “Herr Tristan trete nah” at the very end of the Scene IV. The title actually contains the whole MB work: it’s a fragment which takes literally less than 10 seconds to perform, where Isolde sings just the words “Herr Tristan trete nach” and nothing else. Looking in the score available on IMSLP, I do not understand why this fragment is a separate MB work, it is neither musically nor logically separated from the previous MB work. The only explanation I could imagine that in some releases this fragment is at the end of a recording, in other releases at the start of another recording (I have such release on a CD), and some MB editor decided to create a separate MB work to prevent partial recordings.

  • Fine-grained MB works either became neglected, or require huge ongoing maintenance efforts. Each time a new release is added, it may be required to add new fine-grained MB works because the editor of this release decided on a new split point between recordings. Either new fine-grained MB works must be linked to recordings of all previously existing releases, or those releases will contain wrong meta-information (missing link to a MB work).

  • I do not understand added value of many fine-grained MB works which are never recorded separately. In case of well-known fragments which are often recorded as a stand-alone pieces, such as the “Flight of the valkyries” or famous arias, a separate MB work with an “excerpt” relationship supported by MB still may be used. To a possible answer “fine-grained MB works allows to understand which fragment of a coarse-grained work is on a particular recording” I will argue that this information is available from a name of the recording itself.

So what are the arguments for keeping fine-grained MB works in the MB database?

Well, I can think of a couple of structures that might work.

First, note that both Works and Recordings have duration. They have a starting point, an ending point, and are continuous in between. And note that for Works and Recordings, the end point is knowable. The score defines the end point of a Work. The recorded signal defines the end of a Recording.

Second, note that a Relationship from Recording to Work does not specify a starting point or an ending point. If the Relationship has the “partial” attribute, we have to interpret it as applying from some or all of the Recording, and some or all of the Work.

  1. We could have a structure which represents a point in time within the Work. This could identify the musical moment where a line of words starts to be sung. Or, the start of a musical phrase which leads into a line of words being sung. Or, the end of a section of the music. Or, an arbitrary moment in the music.

  2. We could have a structure which represents a moment in time within the Recording. As with the Work, it could represent a moment when the accompanist starts a musical phrase, or the singer starts to sing a line of words, or the performer ends a section of the music, or some arbitrary moment.

  3. We could have a structure which relates a Recording to a single moment within the Work. It could mean, this Recording entity includes this Moment within the composition.

  4. We could have a structure which relates an interval of time within the Work, with a moment of beginning and and a moment of ending, to an interval of time within the Recording, again with a moment of beginning and of ending. This could mean, the stated interval of time within the Recording corresponds to the interval of time within the Work.

  5. We could specify that the structures with intervals of time are not unique, that there can be multiple such structures, overlapping, mapping from Recordings to the same Work entity.

The tough part is in thinking through the consequences of each of these choices. And, as I mentioned before, establishing use cases and value propositions to justify choosing one or the other.

I understand now! I must confess to being a pragmatist (or at least an aspiring one). For recordings which are “partial”, I tend to rely on the title of the recording to indicate what portion of the work it is a recording of.

I do not like and thus do not edit Wagner :wink:
In a more generic way I’d say: follow the score. BUT I’m very sympathetic with “do not apply any uniform splitting schema”
See e.g. https://musicbrainz.org/work/7ca9655b-8d9e-48be-9917-1e95bf9e183c on which I’m working currently and which tends to drive me crazy :slight_smile:

I only have few Wagner pieces, but the issue may occur at a lesser degree for other composers.

A change of singer could also be considered as a valid work segmentation indicated by the composer. It is actually written on the score. So both “Split by Acts and Scenes”, and “Split by performer change” are valid as they reflect actual segmentation of the work written by the composer. The second option is actually a finer segmentation of the first one. They are not incompatible provided that the singers parts are actually grouped into scenes: Opera > Acts > Scenes > Singer Parts

So, if an opera has been split into “Singer Parts”, we should keep this segmentation, just making sure that the parts are properly grouped into Scenes.

The split into Scenes should just be considered as the minimum required segmentation of an opera. The split into “Singer Parts” should be optional.

So we should focus on fixing works which have been split according to recordings segmentation.

Splitting by “Singer parts” works if such parts are relatively large. In particular, if a work consists of clearly distinct arias, duets, recitatives etc, this approach works as charm. This approach is already used for composers like Handel or Telemann.

With a typical Wagner structure this approach fails. There are a lot of places where many persons (sometimes up to 10) exchange very short replicas. Just look in the libretto how Rienzi starts. If you split each time when a different person speaks, you will get hundreds of micro-works, with a typical performance of each one around 5 to 20 seconds. That is not practical. If you split not on each new singer but only sometimes, we are back to an arbitrary approach where two editors may come to different conclusions. Look above on my example of a currently existing work 10 seconds long.

I like much more proposal from @Jim_DeLaHunt about relative references (possibility to reference a particular part of a work or of a recording), but probably it requires significant changes in MB code and thus is either a long-term solution, or will be never implemented.

Returning to the topic from January regarding a proper way to split Wagner’s operas on parts.

I have cleaned up splitting on parts for all Wagner’s operas except two, problematic cases described below. For all operas except the two problematic cases I applied the approach chosen by the vote above in this topic: use parts explicitly indicated by the composer in the score. I run through the score available on IMSLP and consequently applied splitting chosen by the composer.

Problematic cases:

Das Rheingold: in the score there are just 4 Scenes (no Acts). Current splitting on MusicBrainz in addition includes instrumental interludes (“Orchesterzwischenspiel”) as separate work parts. On one hand, these interludes are not directly indicated in the score. On the other hand, interludes are clearly distinct from Scenes proper, since they do not contain any vocal parts, so keeping them apart is not artificial, is clear to every editor and does not depend on a particular release. I am torn between keeping these interludes as separate works, and merging them with Scenes to keep only parts indicated by Wagner himself. What would you recommend?

Parsifal: Wagner had not provided any parts smaller than an Act, there is no finer sub-division in the score. Acts are pretty long, each of them is typically performed in 1.5 to 2 hours. Although there is no finer sub-division provided by the composer himself, there is a natural sub-division used, for example, in Wikipedia article on “Parsifal”. Similar to “Das Rheingold”, this subdivision is not provided by the composer, but is natural, clear to every editor and does not depend on a particular release. Would you recommend to use coarse splitting chosen by the composer (only Acts), or to apply a bit more detailed splitting, following Wikipedia?

If it’s not ambiguous, I would use the finer splitting … Wagner is notorious for the length of his works.
However I noticed that the parts of the act are not marked as movements.
I would use that, and also use “These relationships have a specific ordering”

Speaking of movements, I noticed that there are many tempo indication in the score.
Movements are often defined based on tempo indications. (Andante, Adagio, …)
In Wagner score, these are in German, but they could be used to split the Acts.
For vocal parts, the incipit of the lyrics may be used to name the movement, as usual.

However I don’t know whether this split would make sense for Wagner. Just an idea after looking quickly at a score. It may be too fine grained …