Gyūdon (牛丼, beef-over-rice bowl) and tonjiru (豚汁, miso soup with pork and vegetables) have much in common. Not only are they a nice combo for a quick lunch (and that it’s almost noon while I’m writing this), but on closer inspection both terms also turn out to be a little off with regard to how they are pronounced.
In Japanese, most kanji have at least two readings. One is derived from its original Chinese pronunciation and called on-yomi (音読み). The other is the native reading that most characters ended up with some time after their import from the continent. This one is called kun-yomi (訓読み). Now when two characters are put together into a compound, called jukugo (熟語), the basic rule of thumb is that the term as a whole should be read either in native kun or in Sino-Japanese on fashion.
To remain in the domain of food for a moment, gohan (rice) consists of two characters: 御 (go) and 飯 (han), both prounounced in on-yomi. Tsukemono (漬物, pickles), on the other hand, is an example of the kun + kun pattern.
The good news is that the large majority of all kanji compounds — edible or not — follow a basic rule: They’re either kun + kun or on + on. This is certainly most appreciated by non-native readers of Japanese (including this author). But unfortunately, as is the case with language, there are a couple of exceptions that make the whole affair a little bit nasty: mixed readings that come as kun + on pairs, or on + kun pairs.
There is even a name for this sort of occasional mating behavior. The former pattern is called yutō-yomi (湯桶読み), which literally means something like “hot liquid container reading,” and the latter, no less enigmatically, goes by the name of jūbako-yomi (重箱読み), or “multi-tiered food box reading.” Sounds kind of weird, but the simple reason behind this terminology is that the two expressions yutō and jūbako are themselves examples of the respective pattern they stand for: yutō is a kun + on reading, and jūbako is on + kun (I’ll put the on-yomi in uppercase to make things easier).
The two names are somewhat misleading though, because both yutō and jūbako are relatively uncommon things in everyday life, whereas mixed readings are definitely not. In fact they are so common that both native speakers and non-native speakers beyond a certain level of proficiency are entirely unaware of them. One very common word that follows the yutō pattern is baSHO (場所, place), which is neither pronounced badokoro nor JōSHO. Likewise, a ticket is called kipPU (切符) not SEPPU, and even if JIKEI would be more logical, a clock is still a toKEI (時計). Two items from the domain of food are butaNIKU (豚肉, pork) and toriNIKU (鶏肉, chicken).
A frequently quoted example of the reverse pattern, on + kun, is DAIdokoro (台所, kitchen). It’s not a DAISHO, nor should it be called utenadokoro. Also, a TV program is not called BANSO, but BANgumi (番組), and each day of the week is a Yōbi (曜日), not a YōJITSU. Our two lunch items, GYūdon and TONjiru (豚汁) fall into this category, too. Strange enough that the most important TONjiru ingredient, butaNIKU, is wrong just the other way round.
The whole affair looks pretty confusing, but it’s not entirely without logic. For one thing, mixed readings may be a good way to avoid homonyms. The term niMOTSU (荷物, baggage), for instance, if read as on + on would become KAMOTSU and thus end up with the same pronunciation as 貨物 (freight). Kun + kun would not be an option either, because nimono (煮物, food boiled and seasoned) is already taken, too. Or take nishiMON (西門, western gate), which would cause much confusion if it were SEIMON, as this is also the term for the main gate (正門).
Another reason for mixed compounds is that in some cases the kun and on reading of a kanji are not interchangeable because they have slightly different meanings: CHū (中) just isn’t the same as naka (中), for instance, and that’s why CHūbi (中火, medium heat) couldn’t become nakabi. Likewise, HON (本) can mean both “origin” and “book”, but moto (本) denotes only the former of the two. You buy a book at a HONya (本屋), not a motoya.
Finally, it may be the case that a kanji has only one reading. BAN (晩, evening) and SHIKI (式, style) are two such examples. Since for some reason they don’t have a pronunciation in kun at all, their on reading is the only option. This may in part explain why we get asaBAN (朝晩, morning and evening) and kabuSHIKI (株式, joint-stock) — though it still leaves us at some loss to explain why the terms didn’t become CHōBAN and SHUSHIKI instead.
Likewise, maybe the only proper way to explain why it has to be GYūdon and TONjiru rather than, say, ushidon and TONJū is that dishes by these latter names certainly wouldn’t taste as good.