inkhaven 1.3.13

Inkhaven — TUI literary work editor for Typst books
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137

# Tutorial 73 — Writing in any language

*Inkhaven 1.3.13+*

Inkhaven's craft tooling — the inline style overlays, the world-consistency
scans, the drift coreference — is **language-aware**. It keys off your project
`language` field, and it works far beyond English. This tutorial shows what's
automatic, what's curated, and how to switch on a language Inkhaven doesn't yet
ship word lists for.

## Set your language

One top-level field in `inkhaven.hjson`:

```
language: russian
```

That single setting steers stemming, the style detectors, the drift pronoun
sets, and the **language the AI world-checks write their findings in**.

## Find out where you stand: `lang status`

```
inkhaven lang status
```

prints an honest coverage matrix for your project language (or pass
`--language <lang>` to check another):

```
inkhaven lang status · language: russian

  stemming                     ✓ Snowball (russian)
  filter words                 built-in 22
  show-don't-tell              built-in 71
  repeated-phrase stop-words   built-in 31
  drift pronouns (coref)       ✓ available
  anachronism lexicon          English built-ins + your `terms` (language-neutral)
  embeddings                   multilingual · MultilingualE5Small
  AI world-check output        forced in russian
  AI world-check prompts       ✓ localized (facts check / scan · drift · continuity)
```

Five languages ship fully curated: **English, Russian, French, German,
Spanish**. For those, every row is green out of the box.

## Honest fallback — no wrong-language flags

If you set a language Inkhaven *doesn't* curate (say `polish`), the detector
word lists come back **empty** — not English. That's deliberate: an empty list
gives you *no* false flags, where the old behaviour would have underlined
English filter-words in your Polish prose. Stemming (Snowball covers 18
languages), the multilingual embedding model, and the AI output language still
work; only the curated word lists are blank, and `lang status` says so:

```
  ▶ no curated detector lists for polish — run `inkhaven lang bootstrap polish`
    or add lists to inkhaven.hjson (stemming, prompts, and embeddings already work).
```

## Curate a new language: `lang bootstrap`

Let the model do the lexicography:

```
inkhaven lang bootstrap polish
```

It asks your LLM — as a careful lexicographer, answering **in Polish**, in
lemma (dictionary) form so Snowball stemming catches every inflection — for the
full detector vocabulary: filter words, the four show-don't-tell categories
(linking verbs, emotion adjectives, manner adverbs, cognition verbs),
repeated-phrase stop-words, and the three pronoun sets (character / place /
artefact). It prints a ready-to-paste HJSON snippet.

Happy with it? Patch the config automatically:

```
inkhaven lang bootstrap polish --yes
```

`--yes` writes the lists into `inkhaven.hjson` in place — with a versioned
backup first and your comments preserved. Re-run `inkhaven lang status polish`
to confirm everything's green, then tune the lists by hand whenever you like.

## The config shape

Bootstrap writes into per-language **maps** — you can also hand-author them:

```
editor: { style_warnings: {
  filter_words:    { languages: { polish: [ "tylko", "naprawdę", "bardzo", … ] } }
  show_dont_tell:  { languages: { polish: {
    linking_verbs:      [ … ]
    emotion_adjectives: [ … ]
    manner_adverbs:     [ … ]
    cognition_verbs:    [ … ]
  } } }
  repeated_phrases:{ languages: { polish: [ "i", "w", "na", "że", … ] } }
} }

drift: { pronouns: { polish: {
  character: [ "on", "ona", "oni", … ]
  place:     [ "tam", "tu", … ]
  artefact:  [ "to", "ten", … ]
} } }
```

Two complete worked examples ship in
[`custom_languages/`](../../custom_languages/): **Arabic** (right-to-left,
pro-drop, suffix pronouns) and **Hungarian** (agglutinative, no gendered third
person). Copy one as a starting point.

## The AI world-checks speak your language too

When you run the world-consistency scans (`inkhaven world --deep`, or
`Ctrl+V Shift+F` in the editor — see [Tutorial 72](72-deep-refresh.md)), the
system prompts for all four scans are **localized** to your language for the
five curated ones, and every finding's explanation is **written in your
language**. An uncurated language falls back to the English prompt with a
warning — never silently.

Want to override a prompt? There's a 3-tier cascade: a `Prompts`-book entry
(`facts-check-ru`, then `facts-check`) → `prompts.hjson` → the localized
built-in. See [`../PROMPTS.md`](../PROMPTS.md).

## Recap

- `language: <lang>` is the one switch that steers everything.
- `inkhaven lang status` tells you, honestly, what's on.
- Uncurated languages get **no** false flags, and stemming / embeddings /
  AI-output / localized-or-fallback prompts already work.
- `inkhaven lang bootstrap <lang>` curates the word lists for any language in
  one command; `--yes` patches the config safely.
- "Does it work in Russian?" — yes, fully; and now in Polish, Arabic,
  Hungarian, or anything you bootstrap.