Bulgarian Crossword Puzzles

These are fully checked grids in the style of the New York Times: every letter belongs to both an across and a down answer, a number sits in the first square of each answer, and the clues come in two lists. A Bulgarian кръстословица lists its clues under Водоравно and Отвесно rather than Across and Down, but the dense, numbered grid is the one Bulgarian papers have run for a century, so nothing about the shape will look foreign to a Sofia solver.

Bulgarian is unusually well suited to it. A dense grid spends about a quarter of its answers at three letters, and most languages built here had to scrape for them — the Greek word list holds 464 three-letter entries, the Catalan one a few hundred usable ones. The Bulgarian list holds 6,681, because the language keeps its short words short: ПЪТ (road), СЪН (sleep), МЪЖ (man), ДОМ (home), ЛЪВ (lion), all with the vowel Ъ doing the work a second syllable would do elsewhere. Even so, the grid did not come easily. The word list's short words are mostly consonant–vowel–consonant, and in a dense corner two three-letter words crossing two others need a vowel where the other word needs a consonant; of the pool's three-letter words only 44 began with a vowel. The two 15×15s and the first 13×13 filled from the most frequent 15% of the vocabulary in under a minute, but the first 11×11 failed 55 layouts at every frequency floor. What filled the six 11×11s was admitting, by hand, some sixty vowel-edged forms the dictionary lists only in its inflection tables — third-person verbs (ПЕЕ, ПИЕ), plurals (ОЧИ, УШИ), pronoun forms (КОЕ, ДВЕ) — and letting three-letter answers recur across the shelf, as ERA and ONE recur in English papers every day. Across the ten puzzles there are 134 different three-letter answers; none appears in more than five, and no answer of four letters or more is used twice.

Every answer is the dictionary form. Bulgarian puts its definite article on the end of the word — КЪЩА is a house, КЪЩАТА the house — and the word list carries 7,816 masculine forms in -ЪТ and 32,888 feminine ones in -ТА, but the clued vocabulary these grids fill from was built from dictionary headwords, so you will be asked for КЪЩА and never for КЪЩАТА. Verbs appear the way a Bulgarian dictionary lists them, in the first person singular (ЧЕТА, I read), which is also how a Bulgarian setter clues them; the exception is a few three-letter answers such as ПЕЕ or ПИЕ, third-person forms that every Bulgarian grid leans on and that the clue says are third-person.

The grid uses the thirty letters of the Bulgarian alphabet: Russian's minus Ё, Ы and Э. The letter that makes Bulgarian look Bulgarian is Ъ, which in Russian is a silent hard sign and in Bulgarian is a full vowel, the sound in БЪЛГАРИЯ itself. It appears in 86,134 entries of the word list and in the most common three-letter words in the language — ВЪВ, КЪМ, СЪМ, ПЪТ, СЪН — so it will cross something in nearly every puzzle. Ь survives only in the sequence ЬО, softening the consonant before it (АКТЬОР, СИНЬО, ШОФЬОР): 4,252 entries, almost all of them loanwords.

Print sometimes marks a stressed vowel with a grave accent to tell two words apart — ѝ (her) from и (and), пàра (steam) from парà (coin) — but the mark is not a letter, and the grid drops it: type И for ѝ. Й, which is a letter, stays; it appears in 45,409 entries.

Bulgarian Cyrillic is outside the one-byte Latin encoding the .puz format uses, so .puz export is switched off for Bulgarian puzzles. Printing to PDF and sharing the puzzle link both work.

The Bulgarian word list combines the bgOffice spelling dictionary — 78,233 stems with the endings they take — with every form that occurs at least twice in Bulgarian film subtitles, and ranks the result by how often each form appears across the subtitles, a 300,000-sentence Bulgarian news corpus and a million sentences of Bulgarian Wikipedia. That gives 593,758 entries. The list is thickest at eight letters, with 87,924 entries, and still holds 6,375 at fifteen; auto-fill takes the most frequent words first and works down the list only when nothing common fits.

The clues start from two Wiktionaries. The Bulgarian Wiktionary, via DBnary, defines about 4,700 of the grid's words — roughly two-thirds of its entries are still the placeholder «Значението на думата все още не е въведено», and those were dropped before anything else. The larger source is the English Wiktionary's Bulgarian entries: 9,583 clues were written from scratch, in Bulgarian, for words that Wiktionary defines in English, using the definitions as a guide rather than translating them. The 546 clues in the ten puzzles themselves were written last, one puzzle at a time, with the whole grid in view.