From 7955fc85ac089d856b44bdea78eccd26ffbd690c Mon Sep 17 00:00:00 2001 From: StefanVukovic99 Date: Mon, 3 Jun 2024 19:25:51 +0200 Subject: display inflection rule descriptions (#1000) * load descriptions in deinflector * description functions in deinflectors * show descriptions in title * use toaster * use names without internal * css lint * reformat transform descriptors * fix merge errors * done? * rename method --- ext/js/language/sga/old-irish-transforms.js | 36 ++++++++++++++--------------- 1 file changed, 18 insertions(+), 18 deletions(-) (limited to 'ext/js/language/sga/old-irish-transforms.js') diff --git a/ext/js/language/sga/old-irish-transforms.js b/ext/js/language/sga/old-irish-transforms.js index d444af1f..793c39a8 100644 --- a/ext/js/language/sga/old-irish-transforms.js +++ b/ext/js/language/sga/old-irish-transforms.js @@ -15,7 +15,7 @@ * along with this program. If not, see . */ -import {suffixInflection, prefixInflection} from '../language-transforms.js'; +import {prefixInflection, suffixInflection} from '../language-transforms.js'; /** * @param {boolean} notBeginning @@ -40,36 +40,36 @@ function tryAlternateOrthography(notBeginning, originalOrthography, alternateOrt export const oldIrishTransforms = { language: 'sga', conditions: {}, - transforms: [ - { + transforms: { + 'nd for nn': { name: 'nd for nn', description: 'nd for nn', rules: [ suffixInflection('nd', 'nn', [], []), ], }, - { + 'cg for c': { name: 'cg for c', description: 'cg for c', rules: [ tryAlternateOrthography(false, 'cg', 'c', [], []), ], }, - { + 'td for t': { name: 'td for t', description: 'td for t', rules: [ tryAlternateOrthography(false, 'td', 't', [], []), ], }, - { + 'pb for p': { name: 'pb for p', description: 'pb for p', rules: [ tryAlternateOrthography(false, 'pb', 'p', [], []), ], }, - { + 'ǽ/æ for é': { name: 'ǽ/æ for é', description: 'ǽ/æ for é', rules: [ @@ -77,7 +77,7 @@ export const oldIrishTransforms = { tryAlternateOrthography(false, 'æ', 'é', [], []), ], }, - { + 'doubled vowel': { name: 'doubled vowel', description: 'Doubled Vowel', rules: [ @@ -88,7 +88,7 @@ export const oldIrishTransforms = { tryAlternateOrthography(true, 'uu', 'ú', [], []), ], }, - { + 'doubled consonant': { name: 'doubled consonant', description: 'Doubled Consonant', rules: [ @@ -105,7 +105,7 @@ export const oldIrishTransforms = { tryAlternateOrthography(true, 'ss', 's', [], []), ], }, - { + 'lenited': { name: 'lenited', description: 'Non-Beginning Lenition', rules: [ @@ -114,7 +114,7 @@ export const oldIrishTransforms = { tryAlternateOrthography(true, 'th', 't', [], []), ], }, - { + 'lenited (Middle Irish)': { name: 'lenited (Middle Irish)', description: 'Non-Beginning Lenition (Middle Irish)', rules: [ @@ -123,7 +123,7 @@ export const oldIrishTransforms = { tryAlternateOrthography(true, 'dh', 'd', [], []), ], }, - { + '[IM] nasalized': { name: '[IM] nasalized', description: 'Nasalized Word', rules: [ @@ -134,7 +134,7 @@ export const oldIrishTransforms = { prefixInflection('m-', '', [], []), ], }, - { + '[IM] nasalized (Middle Irish)': { name: '[IM] nasalized (Middle Irish)', description: 'Nasalized Word (Middle Irish)', rules: [ @@ -143,7 +143,7 @@ export const oldIrishTransforms = { prefixInflection('dt', 'd', [], []), ], }, - { + '[IM] lenited': { name: '[IM] lenited', description: 'Lenited Word', rules: [ @@ -152,7 +152,7 @@ export const oldIrishTransforms = { prefixInflection('th', 't', [], []), ], }, - { + '[IM] lenited (Middle Irish)': { name: '[IM] lenited (Middle Irish)', description: 'Lenited Word (Middle Irish)', rules: [ @@ -161,7 +161,7 @@ export const oldIrishTransforms = { prefixInflection('dh', 'd', [], []), ], }, - { + '[IM] aspirated': { name: '[IM] aspirated', description: 'Aspirated Word', rules: [ @@ -173,7 +173,7 @@ export const oldIrishTransforms = { prefixInflection('h-', '', [], []), ], }, - { + '[IM] geminated': { name: '[IM] geminated', description: 'Geminated Word', rules: [ @@ -201,5 +201,5 @@ export const oldIrishTransforms = { prefixInflection('s-s', 's', [], []), ], }, - ], + }, }; -- cgit v1.2.3