apriiori
random

bingos are difficult in IPA scrabble

IPA English Scrabble is a Scrabble variant / language where, while plays still consist of English words, they consist of their pronunciations written out in the International Phonetic Alphabet instead of their spellings. I first set up a physical copy of this game a few years ago; I am currently setting up an online version. You can play it here, though note that it is currently in beta (i.e. unlisted on apriiori.com) as I am still making balance changes and refinements to the word list.

A few tileset proposals for IPA English Scrabble can be found on Wikipedia. They typically include the tiles: ə, n, ɪ, l, s, t, k, r, d, i, m, ɛ, z, ɑ, æ, b, ow, p, aj, ej, f, g, ɔ, v, h, ŋ, ʃ, u, w, dʒ, j, tʃ, aw, ɔj, θ, ʊ, ð, and ʒ1. This list targets General American; one would use a slightly different phoneme set for, say, British English, and probably also change the tile frequencies around a little. The standard list has the father-bother merger.

While I find it unintuitive due to not myself having the merger, I think one also wants to cut the ɔ tile and play the game with a cot-caught merger. For one thing, many General American speakers have the merger and won’t be able to particularly distinguish when to use an ɔ or a ɑ, but also, in terms of gameplay balance — the tileset already sort of has a lot of tiles, certainly substantially more than 26. This means it’s desirable to cut tiles when possible, and ɔ is the most cuttable tile.

On the other hand, the set I currently use introduces the ɚ tile. I think this takes some load off the ə tile, and writing it as ər feels really weird. You could, one supposes, write it as a syllabic r, but that feels sort of weird too. That’s probably one of the questions I’ll revisit.

The fundamental difficulty of balancing IPA English Scrabble is that an English word typically has substantially more letters than it has actual phonemes. That is to say, a given English word usually consists of substantially fewer tiles in IPA than in the standard English orthography. This leads to plays in IPA English Scrabble tilting a little towards the shorter end. You see more three letter plays or four letter plays, and bingos become substantially harder to put together. Some of that effect is likely people not being used to IPA quite as much as the normal orthography, even when they are fluent in IPA and certainly when they aren’t; some of it is also probably worse intuitions for how to go about playing IPA English Scrabble as a game, what sorts of leaves are good, etc. But some of it is also probably inherent to the game.

There are a few approaches you can take here, none of which I’m sure I’m happy with. You can fiddle with the tile distribution a little, but none seems to do much substantially better all-around than basing it on letter frequency. The English tileset differs from letter frequency mostly in dramatically cutting the s tiles and thus making plurals (and third person verb forms) a little less readily available, giving some extra tiles to be spread around other letters; this doesn't quite work so well in IPA, because the voicing split between s and z means half the plurals take one tile and half take another tile, so they’re already a lot more constrained without cutting any tiles below the amount suggested by their frequency. One thing I’m trying is having an s/z tile, a sort of semi-blank tile that’s restricted to those two options — this makes bingos somewhat more tractable, but doesn’t solve the whole problem, and also feels a little weird.

Another option currently available on the website is to give a half-bingo bonus for a six tile play. I’m not certain how I feel about that, but it’s an option. Mostly I’m still thinking about ways to resolve this — I suspect a good solution will probably involve doing something galaxy brain with the tileset, but I’d sort of like to figure out what the least galaxy-brain option is, and also try to tease apart how much this is a fundamental issue with the game balance vs. an issue with no one being particularly practiced at the game yet.

One of the most exciting ideas is to split the diphthongs, i.e. remove all the diphthong tiles, have those sounds be spelt with a pair of single tiles. This would probably help. But figuring out exactly how to transcribe each diphthong that way is a little tricky. The main compensating change to the tileset would probably be the addition of a lot more j and w tiles. I’ll probably try to set up a mode for this next and see how much it helps.

  1. An insane version described by Wikipedia also gives

    1 ʙ worth 14, 1 ʛ worth 12, 1 ħ worth 9, 1 kʼ worth 11, 1 ɮ worth 11, 1 ɲ worth 13, 1 ʉ worth 11, 1 ⱱ worth 11, 1 ǃ worth 16, and 1 ʘ worth 18

    This is the sort of awful bullshit I’d create as a joke, it’s wonderful.

    ↩

26 comments mirrored from Substack

  1. homura | fastfouriertransfem · 1 like

    > 1 ʙ worth 14, 1 ʛ worth 12, 1 ħ worth 9, 1 kʼ worth 11, 1 ɮ worth 11, 1 ɲ worth 13, 1 ʉ worth 11, 1 ⱱ worth 11, 1 ǃ worth 16, and 1 ʘ worth 18

    how are these even used ??

    1. April · 1 like

      I would love to watch even a single game played by whoever set up that tileset. Because I have no clue.

  2. anon · 1 like

    Eric Harshbarger (I think) invented a Scrabble variant where you get to choose to draw to up to 15. Then bingos are exactly plays using your whole rack and the bingo bonus equals the square of the number of letters used.

    1. bingo maxxing by getting one point bonuses for playing two tile words every move

  3. Claude and I trained wolges (https://github.com/andy-k/wolges) to play the IPA variant and it hits 2.18 bingos *total* per game (which is close to i.e. Nigel Richards's *per player* value in English) vs English's 3.81. The split diphthong version gets closer to English's bingo rate.

    I'll probably use the bot to try to better calibrate stuff like tile values.

  4. usablejam · 1 like

    I skimmed this and don't see where you considered having a hand be fewer than seven tiles.

    I did try to see if I could search for "six" or "seven" (¯\_(ツ)_-) to make sure of this, but, although Discord allows me to "scan for AI text" (presumably using a comprehensive simulation of all of history), "search in page" has not yet been implemented.

    1. i think having a hand be six tiles crossed my mind but i haven't tried it

      1. usablejam

        Perhaps having many tokens to juggle makes that a bad idea...

  5. usablejam · 1 like

    > ɚ

    I don't get this one. I'm listening to the "r-colored vowel" recording on Wikipedia, which just sounds like "rrrr".

    1. usablejam · 1 like

      My feeling has been that some non-rhotic speakers don't want to admit that they simply forgot to pronounce the "r" in many words, and so have invented a whole fan-fiction about it.

      1. this is definitely not the case there are predictable phonological rules about where the rs are dropped or inserted. (coda r's aren't pronounced, but an r is inserted whenever hiatus would otherwise occur, i.e. when a word ending in /ə/, /ɪə/, /ɑː/, or /ɔː/ is followed by a word starting with a vowel)

        1. usablejam · 1 like

          Well, sure, there are patterns in where they forgot to pronounce the "r", I agree.

          1. yeah and also ones where you forget the r

            1. usablejam

              I'm not sure what you mean

              1. like the first r in "law and order"

                1. usablejam · 1 like

                  Normal people who still remember how to pronounce their "r"s pronounce both "r"s in "order" the same way.

                  There's a reason they're spelled with the same letter! Before you Brits started dropping some of the sounds, they were the same! Over here in the US, we still pronounce them the old-fashioned-British way: the same!

                  1. April · 1 like

                    no i meant the one between law and and

                    1. usablejam · 1 like

                      Ha, yeah, my grandfather from Brooklyn would have put an "r" there too.

                      "Lawr an' order."

                    2. usablejam · 1 like

                      But like, when did that come about? It must be recent, right? There's a reason the "r" sound there not only isn't universal but isn't even ever written down.

                      1. Wikipedia gives the very helpful and precise "sometime after the year 1700"

                        1. usablejam · 1 like

                          Oh, that reminds me of a good one I saw.

                          I was discussing with ChatGPT a certain controversy involving the historicity of A, and whether it significantly predates year X. ChatGPT checks some sources and reports that A dates in fact all the way back to the late B period. I am surprised, because the B period started hundreds of years before X. Then I notice that the B period lasted until ten or fifteen years after X. Then it turns out that "late B period" in this case actually means "a few years before X, at most..."

        2. usablejam

          This just sounds like, "Oh, no, it's not fan-fiction! Anyway, the rules of my alternate universe are as follows: an 'r' is present when I say it is, even if it's impossible to detect without context; Hermione is a 12' tall male centaur and is dating Sirius; ..."

          1. look it's just how people speak in britain man it's not that complicated

    2. i mean yeah. it's just like rrrr. but if rrrr were a vowel. like in bird, or yearn.

      1. usablejam · 1 like

        If "l" a vowel in "wild" and "kiln"?

        1. maybe in wild no in kiln

Join the discussion on Substack