DCMLab / ABC

Annotated Beethoven Corpus (ABC): A dataset of harmonic analyses with standardized labels
https://www.frontiersin.org/articles/10.3389/fdigh.2018.00016/full
Other
45 stars 15 forks source link

n13op130_03 key change issue, m. 30 #30

Closed malcolmsailor closed 1 year ago

malcolmsailor commented 2 years ago

As you can see, on the downbeat of m. 30, there is a chord with an alternate label, either "IV6" or "I.I6".

The issue is that the alternate label changes the key, and the analysis from that point proceeds in that key, but the harmonies tsv file remains in the other key (which is V).

I'm posting this issue in part because I'm curious what the correct behavior is. When an alt label has a key change, should that key change persist?

Screen Shot 2022-08-09 at 3 25 01 PM

(Another minor note about this extract is that the annotator seems to have neglected the C-flat at the end of m. 30.)

malcolmsailor commented 2 years ago

In fact it seems that there are a number of other such cases. I just made a big table and queried it for all alt_labels that contained the '.' character. Nearly all occur in op130. I looked at the op95 example and in that case the annotator didn't expect the alt_label key to persist. In all the op130 examples I looked at, in contrast, the annotator seems to have expected the op130 label to persist, leading to all the following chord symbols being in the wrong key in the table. For example, in mn 20 of n13op130_01, the key persists as B-flat, rather than moving to the dominant F.

      mn mn_onset         filename alt_label
286  148      3/8   n11op95_02.tsv  ii/.bIII
63    20        0  n13op130_01.tsv       V.I
536  173        0  n13op130_01.tsv     I.i64
73    37        0  n13op130_02.tsv     II.ii
147   30        0  n13op130_03.tsv      I.I6
313   63      1/2  n13op130_03.tsv      IV.I
77    56        0  n13op130_06.tsv       V.I
257  195        0  n13op130_06.tsv       I.I
567  455        0  n13op130_06.tsv     I.ii6
591  470        0  n13op130_06.tsv      IV.I
johentsch commented 2 years ago

The rule has been (since the ABC) that a key change cannot occur (only) in the alternative label. Thanks for looking these up!

The case you show from op95 is not a key change, but an over-generalization of the principle that labels beginning with b, at the time, needed to be preceded by a .. Looks like I should do a cleanup. As you noticed, this will require looking into the score at each point to see what the correct key is.