Change log

1.0.1 broke gulp-rosaenlg. Removed more deeply dependancy with uglify (which was used by Pug pre-RosaeNLG).

Browser compilation was broken. Fix + added automated tests.

search engine available in RosaeNLG doc thanks to Algolia DocSearch (awesome product and support btw)

outdated packages (yarn outdated) replaced by newer version

1.0.0 version at last!

new logo

open-source publication

nothing as it did never exist

nothing as it did never exist

security fixes: 0 vulnerabilities found - Packages audited: 57380

style for == and !=

usage of recast to parse js expressions when checking for linguistic resources

much better support of html when filtering

removed (some) deprecated libs

compare with CoreNLG (to be enriched with the feedback of the CoreNLG team!)

proper capitalization and spacing when mixin with html bold or italic

removed sqlite dependancy (because it was hard to build & not really useful)

adj in value mixin accepts more: can be a list adj:['beau', 'grand'], or an object adj:{ BEFORE: ['beau', 'intelligent'], AFTER: ['brun'] }

DEMONSTRATIVE determiner type in Italian

possessive adjective param in value possessiveAdj for Italian possessives

boilerplate documentation see boilerplate project

renamed for freenlg to rosaenlg

Indefinite articles in German simplified syntax <…​> parsing

Italian verbs

French past participles used as adjectives

Italian past participles and present participles used as adjectives

German present participles used as adjectives

better Yseop templates generation

first support of Italian (no verbs)

basic support on any language

INDEFINITE determiner type in German

Non-breaking space in French punctuation (before : ! ; and ?)

de #[+value(9000)] (or any number) would contract into d'9 000. Now generates de 9000 properly.

Punctuation on French and English when using ? or !.

French possessives issues.

Added dates formatting thanks to moment lib

Filtering can be explicitely disabled with disableFiltering: true in renderFile

Numbers formatting: use FORMAT in value to set a format directly used by numeral. See numeral.js formats. Very practical for currencies, %, etc. For instance +value(104000, {'FORMAT': '0a$'}) will output 104k€ when generating French.

det to add a determiner (French and German); current determiners are DEFINITE and DEMONSTRATIVE.

It is now easier to complete the params object with new values with addToParams: addToParams({xxx:yyy}) is equivalent to Object.assign({}, params, {xxx:yyy}.

adj property in value to add an adjective.

owner property in value to manage possessives. Does the same thing as thirdPossession.

recordSaid and deleteSaid do not need a `- ` before them no more.

Ability to globally choose the best synonymic alternatives with choosebest: see [choosebest].

Ability to compile and/or just render texts in browser, without using node.js. See [inbrowser].

Tutoriels can be run directly in the browser.

Improved French support, see below.

Partial support of German (de_DE), see below.

Ability to generate Yseop templates. Yseop is a NLG software vendor. See [yseop.adoc].

determiners and possessives (thirdPossession)

Some speed optimizations (which are not significant)

Refactoring: switched to TypeScript for most of the code

Some renaming: please impact your templates

Updated librairies and linguistic resources documentation: see index.adoc#resources

Replaced better-title-case, which was deprecated on npm, by better-title-case. Changed the test case (titlecase rules seem to be complex).

monthName(date) (which generated the name of the month january february etc.) is deprecated as we now have moment lib. Use #[+value(date, 'MMMM')] instead.

Removed format-number-french (replaced by numeral).

Removed jslingua (replaced by custom lefff extract).

valWithUnit is deprecated and removed.

loadDict param is removed; ressources a just loaded when necessary.

Removed registerSubst. Use setRefGenderNumber directly.

updated many libs: copyfiles mocha rimraf better-sqlite3 compromise moment written-number

no changes - just fixin Lerna

performance documentation

Sub modules are now referenced via their exact version number (no more ^…​). It will be easier to use fixed versions of RosaeNLG (with their dependencies) in the future.

French support for month names did not work - fixed

getNextRep (used in French) didn’t work properly: the chosen representant was often not the one that getNextRep had predicted.

anaphoras: after forcing referential representant output, the next one was still the representant; now it is the anaphoric one.

mostly refactoring

syno_sentences mixin deprecated, use synz > syn structure

assemble mixin deprecated, use itemz > item structure

global filtering is done automatically unless any filtering was been done during generation via filter mixin

noFilter parameter is deprecated

French support for months

refactoring, no feature change

pug render did not work - fixed.

doc is now also generated in one single large PDF file

to raise the probability of a specific synonymic alternative to be triggered, use syn {weight:4} (4 times higher chances - default is 1)

to force a specific synonym to trigger, use synz {force:3} (to trigger the 3rd one)

synz 'sequence' syntax is not valid no more: use synz {mode:'sequence'}

changed the random algorithm: it is still random but does not provide exactly the same numbers as before; you shall update your regression tests (and I’m sorry for that).

defaultSynoType renamed to defaultSynoMode

setSynoProps, setSize, getSize removed

semicolon (;) support in punctuation (and taking care of HTML entities which contain a semicolon)

spaces are automatically added before and after various mixins

titlecase is now a structure element and no more a mixin:

FYI the titlecase npm package, which is used in en_US, is no longer maintained

eachz structure as (preferred) an alternative to the foreach mixin:

main.pug is included automatically so do not include it yourself explicitely in your templates no nore.

new protect structure tag that acts like protectString, to exclude some text from the filtering process:

path issues to get RosaeNLG’s mixins

align versions between rosaenlg and rosaenlg-core

wrong French resources path

wrong path in include node_modules/rosaenlg/main.pug

easier integration:

automatic filtering unless noFilter is set to true

new boolean parameter noFilter if don’t want to automatically filter the whole outputs

new structure keywords:

a could be transformed into the by compromise in some situations

When listing elements in a foreach or in an assemble, the list of the non empty elements is made available in params.nonEmpty.

a/an specific cases, when a is:

…​ is automatically transformed into …

Proper punctuation around …: no space before and after etc. No automatic capitalisation after …, as it depends of the context.

mixin eatSpace that 'eats' the spaces around him.

a/an when uppercase: A apple now outputs An apple

missing French contractions when uppercase: De les→`Des`, De le→`Du`

in assemble or foreach with paragraph: true: paragraph parameter is deprecated, use either:

in assemblies, when mode is sentences or paragraphs:

a first version of the tutorial Tutorial

in assemble, when the mixin is not set, will automatically try to use value

in assemble and foreach: shuffle was done too late (after evaluation)

filter: remove spaces at the very end of the text

filter: capitalize the very beginning of the text

filter: capitalize after exclamation mark

filter: removes extra ponctuation when "!" and "." are combined

A shy start of French conjugation with jslingua lib. No support for group 3 verbs.

formatting of numbers in French in value thanks to format-number-french lib

plural of words in French: œil ⇒ yeux etc. thanks to pluralize-fr lib

title case (which is more difficult than it appears) in English thanks to titlecase lib, in French thanks to titlecase-french lib

</p> . </p> was not properly filtered

synonym mode is now a global parameter (random - default, or sequence), via defaultSynoType in the constructor ; can be overridden locally using setSynoType; see mixins_ref.adoc#synonyms_mode

a / an transformation was triggered in all languages, fixed to English only

a/an in English: a industry ⇒ an industry using compromise lib

function isNotEmpty() is deprecated because it brought side effects.

bad prediction in assemble and syno_sentences due to test context that was too narrow - see listOfParaTricky in test_assembly_para

uses compromise lib in en_US which was the basis for multiple new features in English:

direct access to compromise lib via util.compromise

empty prediction mechanism improvements:

randomSeed param of NlgLib constructor renamed to forceRandomSeed

list of words with their gender in French

french adjectives agreement rules (well, most of it)

adds spaces in assembly separators automatically if the user does not put them

notion of language: fr_FR or en_US