Word boundary

[user202729]

Word boundary conflict is an issue. Right now AM/I produces [ami] (while the program can generate entries that use suffixes, it so far cannot filter out entries that should use them but doesn't) Similar cases: for a^ (fora), for a (foray), on to

[user202729]

The suffix cases are mostly done now. But there are cases such as (sun shine), (to-do), (goto), (some things), (school bus) or in general compound words.

Non-compound word cases:: toews (to use), morcom (more common), orloff (or love), reidt (read the) (but -T is part of plover theory too).

(TODO download wikipedia or something similar and detect those cases automatically)

[user202729]

Then there's the reverse case (Italian -- prefix i) or battlefield (because -L is suffix and /əɫ/ cannot match separate -L).

[user202729]

With the new word_boundary_conflicts script... it would be good to be able to mark "less favorable strokes" (misstrokes, etc.)

so the compound word option is preferred.

(partially fixed: add "disambiguation_stroke" command-line argument)

[user202729]

Case analysis:

They can be split into two general groups: (do not handle manually anymore, make it automatic, see below)

Large word should be pushed

• "come back", etc.: compound word.
• "to land", "we know", "to use", "by number": mark that WE, TPHOE, TO, AOUS, PWEU, TPHUPL are the main outlines for the words, so the compound word is secondary.
• "an external" (annex): EBGS is a prefix stroke, should not be used at the end of a word.
• "or another" (orran): APB is not really a prefix stroke, but it should not be used at the end of a word; on the other hand, the main stroke for "another" is TPHOER, so it doesn't matter too much.
• "are well": Separate left-hand strokes should be removed if there's any better matching.
• "all or", "such an", (allor, suchan): should force the suffix stroke.

Small word should be pushed (i.e. do nothing)

• "to feature": TPAOEFP for "feature" is quite uncommon. (this word is in Plover dictionary, but without any TO/ prefix.

  TO/ in Plover main dictionary is rare, but not non-existent -- "totalitarian", (not correct -- pronounced with /oʊ/), "tonight" (compound word), "tonofilament" (???), topography, toronto, tobacco (with C consonant doubling) -- in all of those cases they do not form a phrase)

  May tell the user about the brief so they can use that and avoid conflicts.

• [x] "and elsewhere", "in Japan": the main stroke for "and" is SKP and for "in" is TPH, doesn't matter too much.

[user202729]

For now, the large word should always be pushed.

• [x] implement it

The user can choose to mark small word as misstroke/unused so that the large word is not pushed.

13 will be harder in that case, however.

[user202729]

... which means that "small word should be pushed" is the special case.

So the plan is...

• When the user write TO/FEE (translated to "to fee" instead of "toffee" with the feature implemented, the program should
  • Tell the user that
    • TO/FEE maps to "toffee" (can also be written TOF/YI or TO/FEE + disambiguation), and
    • FEE/TUUR maps to "feature" (can also be written FEECh) (and also the other possible cases -- "to female, to FIFA, to Fiji"), and
  • Ask if the user want to: (bold = should be chosen)
    • delete (deprioritize) (TO -> to), or
    • force TO/FEE -> toffee, or
    • delete FEE/TUUR, etc., or
    • delete TO/FEE -> toffee and never show the warning again, or
    • (manually) delete them and write TOF/YI instead.
• (imaginary case, FO/FO is in the (main) dictionary so it will be generated) When the user write FO/FO (does not translate to FoFo with the feature implemented, instead translate to "to to"), similarly, ask if the user want to
  • delete (deprioritize) (FO -> to), or
  • force FO/FO -> FoFo (will be done automatically once option 1 is used)
  • delete FO/LOE -> follow, FO/RUN -> foreign, etc., or
  • delete FO/FO -> FoFo (effectively no-op, except that the warning is never shown again), or
  • (manually) write FO/FO + disambiguation instead to write FoFo. (very bad, especially when there's no alternative stroke)

---

In fact, there's no need to deprioritize the small words (FO, AND); however, warn the user about the possible word boundary issues -- unless it's marked as deprioritized.

• [ ] implement it