Better exact pitch value handling

[eao]

Unless I am misunderstanding, the exact pitch value handling introduced in #2055 does not allow for distinguishing between heiban and odaka, since, for a 3-mora word, both would just be LHH. I am trying to make a pitch dictionary with the new string type pitch but am unable to make it work like the integer type pitch does.

Example of current behavior:

Also, different parts of speech seem to behave differently, as this word seems to have different patterns than that of the 打ち込む example from #2055.

Maybe there could be a way to specify additional downstep(s) outside of the morae of the word itself? The mora after the word ends is the most important, for differentiating heiban and odaka, but supporting a downstep for the mora before a word begins also has uses, like for suffixes.

Example - Daijisen has this data for 歌, depending on whether it is used on its own or after another word, like in 童歌 (わらべ↓うた).

うた↓｛が｝ ／【複】～↓うた

It would be nice if Yomitan could support both these cases, though again, the heiban/odaka case is the more relevant one.

[eao]

Furthermore, it would be good if there were some way of explicitly supporting compound word pitch, like yoji where one word is pattern A and one word is pattern B.

[Kuuuube]

Is the goal here to have control over the "suffix" pitch? This is already possible simply by adding an extra H or L.

For example with おとめ instead of HHH you could do HHHH or HHHL. Shown below is HHH, HHHH, HHHL:

Or LHH, LHHH, LHHL:

For the part of speech, if by pattern you mean category aka the result of {pitch-accent-categories}, indeed that is indended. If you meant anything else, that is not intended and please provide more info.

For compound pitch are you simply looking for some sort of separator to be able to put between the pitches?

[eao]

I was not aware of the ability to add an extra H or L to control the "suffix" pitch, sorry and thank you. I am not sure if/where this is documented; I looked through dictionary-term-meta-bank-v3-schema.json but did not see it. That was my biggest concern, but is not an issue now.

Adding support for "prefix" pitch, a pre-word downstep, like for the example of 歌 when used as a suffix, is pretty niche, but still potentially useful. It could allow someone to make a suffix-based pitch dictionary that would only show up when a word has a unique pitch pattern as a suffix. It would, however, require a different method of indicating a downstep, since we can't just add an extra H/L at the beginning of the word. Maybe something like a "↓" before the H's and L's start? For example, "↓LL" for the suffix usage of 歌. Whether this is considered worthwhile enough to implement is up to debate, but the use case exists.

Additionally, this "↓" notation could also be used at the end of a word for heiban/odaka differentiation instead of an extra H/L - absence of "↓" indicating heiban and presence of "↓" indicating odaka. For example, "LHH" = heiban and "LHH↓" = odaka. Opinions may differ but I think this would be more intuitive, making the numbers of H's + L's directly correspond to the mora count rather than sometimes having an extra one. That said, the differentiation is already possible with the extra H/L so this is not that important.

That leaves the compound pitch. Yes, a separator is definitely desirable, as some resources already have helpful separators. For example, Daijisen, for the yoji "一喜一憂", has pitch info " い↓っき△いちゆう○ ／い↓っき△いち↓ゆう". So either [1-0] or [1-2], with the boundary being between the いっき and いちゆう. While both pitch patterns can already be expressed in the current format (HLLLHHHH and HLLLHLL), it is not clear from the H's and L's alone where the word boundary is. Take the second "HLLLHLL" pattern - even if it is assumed that there is only one word boundary, this could be interpreted correctly as [1-2] "いっき-いちゆう", or, incorrectly as [1-1] "いっきい-ちゆう". Having a separator would make this clear, which would be helpful as-is, and also enable extra Anki functionality like coloring words based on pitch. So, if the separator were a hyphen, "HLL-LHHHH" or "HLL-LHLL".

Some dictionaries' entries for compound words also have information about where the boundary in the expression itself is, but that is probably outside of the scope of pitch representation.

[Kuuuube]

Indeed the suffix pitch was not in the docs. PR made for that #2298.