In English (I haven't really thought too much about English yet), there are tons of what-seem-like one-off patterns.

(the "oo" sound)
tool /tul/
two /tu/
to /tu/
through /θɹu/
blue /blu/
queue /ku/
(highly variable)

(the "uu" sound)
book /bʊk/
(pretty consistent)

(the "eh" sound)
bed /bɛd/
spread /sprɛd/

Then you things like (i, ai, ae) being the same sounds, or (ie, ee). Etc.

Is there a reasonably small number of these combinations in English? Like, are there rules for all of them, or are there a bunch of exceptions? Like aardvark. (hour, our) (own, hone, moan), etc.

I ask because I'm looking at different languages and (at least as I am first starting) I am seeing that the sounds and vowel symbols have 1 or perhaps 2 sounds depending on a small amount of context. Either it's a different dialect, or the position of the letter relative to something else (like sound A for letter x after a consonant, sound B after another vowel) (or sound A at the end of a word, sound B everywhere else). It's appearing that no language is as complex as english in the number of sounds per letter sequence, and number of patterns and variations like I started to list above.

Which languages are similar to English in having a largish number of rules for letter sequences and what sound they produce? Are all/any/some/many languages strict with the number of combinations, limiting it to a 1-to-1 mapping? Does it just depend, and there's not a finite set of rules you can map out? Does English have a large number of combinations (in the 1000's +) or is it relatively small (100 or so).

For instance, Japanese and Tibetan and Sanskrit all seem regular in that there is a 1-to-1 mapping from sound(s) to letter(s). I haven't learned too much about each yet though so I'm sure there's edge cases (and it would be interesting to know).

  • It's definitely untrue that Japanese has only one sound per "letter": almost every kanji can be pronounced in more than one way (often very different ways), and even kana has ambiguity, and had a lot more ambiguity before it was reformed after the war (for instance, きょう kyō was written けふ kefu if I'm not mistaken). – LjL Sep 29 '19 at 14:32

To rephrase the question, it is possibly hardest to go from pronunciation to letter sequences in English, compared to other written languages. I might propose Chinese as being worse (even less predictable) except I assume you only mean alphabetic-like systems. Lhasa Tibetan and Mongolian in Uighur script might be close runners-up. English uses a combination of rules and strategies, for example [f] is most likely spelled ph in words of Greek origin, [i] → ae in Latinate words but ee, ea in Old English words. To the extent that you might guess the source of a word, that helps you predict the spelling (also any related words such as "paradigm; paradigmatic"). I have never seen anything that resembles an exhaustive "set of rules" for pronunciation-to-spelling mapping in English. There may be some core that they teach in schools, and the rest is left to memorization.

Since most languages have writing systems of more recent provenance or had an orthographic reform, most languages are not as challenging in the pronunciation-to-writing mapping.

  • That "core" taught in schools is called phonics. – Nardog Sep 29 '19 at 1:48
  • I know of it, I'm old. Do you know if there is a standardized set of rules that constitute the curriculum? – user6726 Sep 29 '19 at 1:53
  • My computational linguist friends tell me that building computer systems that do text->speech requires around 2,000 rules of mappings between written and spoken English. So there are potentially that number of 'phonics' rules, but of course no-one teaches anywhere near that number, kids figure most of them out. – Gaston Ümlaut Sep 30 '19 at 6:16

It sounds like you're asking for orthographic rules, a set of rules that can convert a written form (a series of graphemes) to an underlying phonological form (a series of phonemes).

For most languages with modern orthographies, the number of rules is relatively small. They don't all follow the principle of one-grapheme-per-phoneme, which is considered pretty much ideal nowadays, but there tends to be an unambiguous or almost-unambiguous way to map from one to the other with consistent, predictable rules. (For example, such a mapping for Swahili would require somewhere between 10-100 rules; traditionally, these mappings are implemented with finite state transducers, so people talk about "states" instead of "rules", but the idea is the same.)

On the opposite end of the scale, English, Mandarin, Japanese, and a few others throw that principle out the window. It's possible to do reasonably with English; this attempt uses ~100 rules…and ends up with 59% accuracy on the same corpus used for training. If you want to do better than that, or have any success at all with Han-based orthographies, you'll end up compiling a huge lookup table of one-off rules to tell you that 山 means /yama/ and 水 means /mizu/. This will most likely require something between 1,000-10,000 rules, for a conservative estimate.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.