Change log of RosaeNLG

This is the documentation for 1.5.7 version, which is not the latest version. Consider upgrading to 2.0.0.

[1.5.7] - 2020-01-05

Changed

  • removed snyk dependancy

[1.5.6] - 2019-12-14

1.5.5 was not a really good version - please favor 1.5.6 instead.

Fixed

  • quotes ("…​") are now properly managed in terms of spacing

Changed

  • removed filter mixin: everything is filtered by default; use protect mixin to protect sections you don’t want to filter

  • removed disableFiltering option

[1.5.5] - 2019-12-13

Added

  • in rosaenlg-node-server: ability to render a template directly, by sending the template and the data in the same request

Fixed

  • the "missing space" issue, which lead to add lines with just | or | #{''}, is fixed (which is indeed a very good thing)

Changed

  • §…​§ structure is deprecated: please use protect tag instead which better cares for spacing

  • in rosaenlg-node-server: removed counter in the response (didn’t work properly, and was not really interesting)

[1.5.4] - 2019-12-04

Added

  • new additional independant module content-deduplicate; it is not used by RosaeNLG while rendering texts but you can use it afterwards to assess the quality of your texts

Changed

  • in synonym global optimisation: html tags div p etc. are not considered no more as words; this can bring regressions on your projects, as synonyms are better chosen

  • French is better tokenized when there are apostrophes in the text

Fixed

  • dependancies update

[1.5.3] - 2019-11-24

Added

  • list of standard stop words in the documentation

  • Developer experience documentation

  • with CLI, templates can now be rendered in stdout (not systematically in a file)

  • compare with xSpin

  • examples of templates showing the added value of a Natural Language Generator compared to a plain template engine like Pug

Fixed

  • itemz and eachz can be used with parameters: default will be {}

Changed

  • rosaenlg-cli used to output in files by default; now by default it outputs in stdout; add --out=./ to output in files as before

[1.5.2] - 2019-11-18

Changed

  • docker images (command line interface and server) are not built through main CI and available here

  • browser-ide-demo is now a package of the project

[1.5.1] - 2019-11-16

Fixed

  • rosaenlg-node-server was not able to find openApiDocumentation_merged.json as path was not relative

[1.5.0] - 2019-11-16

Added

Changed

  • rosaengl-cli (the command line interface):

    • is now in the main repo

    • updated dependancies on rosaengl-cli

  • rosaengl-node-server (server based on RosaeNLG):

    • is now in the main repo

    • logs in rosaengl-node-server

Fixed

  • removed @types/random-js dependancy which made a warning

Changed

[1.4.1] - 2019-11-15

Added

  • NlgLib is now exposed. Useful when using compileFile (e.g. import {compileFile, NlgLib} from 'rosaenlg'),and mandatory for rosaenlg-node-server.

Changed

  • update of many dependancies

[1.4.0] - 2019-11-07

Changed

  • large reorganization of the documentation

Added

  • more doc on how to integrate, Java Wrapper, Java Server etc.

[1.3.3] - 2019-10-29

Fixed

  • #[+value('some string')] didn’t work no more

[1.3.2] - 2019-10-27

Added

  • in the browser files supporting compilation (_comp.js), there was only the render function; added compile, compileFile, compileClient, compileFileClient, renderFile and an access to NlgLib

  • in gulp-rosaenlg, packageTemplateJson to bundle templates and included templates in a single Json file

Changed

  • removed 'fat js' version (was huge, long to produce & not really useful)

[1.3.1] - 2019-10-18

Fixed

  • fat js is still as fat but at least works now.

[1.3.0] - 2019-10-13

Added

  • fat js packaged version, with compilation ability on each language, to be used in GraalVM

[1.2.0] - 2019-10-12

Added

  • recordSaid hasSaid and deleteSaid existed to safely record booleans; now there is recordValue getValue and deleteValue to safely record numbers or strings

  • compare with CoreNLG in the documentation

[1.1.1] - 2019-10-09

Added

  • GraalVM runtime documentation

Fixed

  • Dev dependancies security issues thanks to Snyk

  • More French contractions (j’ai, lorsqu’il, puisqu’elle, jusqu’à etc.) thanks to CoreNLG source code

Changed

[1.1.0] - 2019-10-07

Added

  • new synonym mode once: will trigger each alternative randomly, but will try to trigger each one only once. Just use it instead of random!

[1.0.5] - 2019-09-29

Fixed

  • README.md is back for npmjs.com

  • in French, on browser package, composed tenses raised an error, because verbs that always take ETRE aux were not correctly packaged

[1.0.4] - 2019-09-29

Changed

  • reduced size on npm thanks to .npmignore

[1.0.3] - 2019-09-29

Added

  • when using subjectVerb in French, when aux is ETRE, the participle is automatically agreed with the subject

[1.0.2] - 2019-09-12

Fixed

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

[1.0.1] - 2019-09-12

Fixed

  • Browser compilation was broken. Fix + added automated tests.

[1.0.0] - 2019-09-10

Added

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

Fixed

  • outdated packages (yarn outdated) replaced by newer version

Changed

  • 1.0.0 version at last!

[0.18.10] - 2019-09-07

  • doc update

  • search engine in the doc (thanks to Algolia)

[0.18.9] - 2019-09-05

Changed

  • new logo

[0.18.8] - 2019-09-05

Changed

  • open-source publication

[0.18.7]

Changed

  • nothing as it did never exist

[0.18.6]

Changed

  • nothing as it did never exist

[0.18.5] - 2019-08-27 (private version)

Fixed

  • security fixes: 0 vulnerabilities found - Packages audited: 57380

    • updated gulp-pegjs version

    • new version of libs

[0.18.4] - 2019-08-27 (private version)

Changed

  • style for == and !=

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

[0.18.3] - 2019-08-02 (private version)

Fixed

  • much better support of html when filtering

  • removed (some) deprecated libs

[0.18.2] - 2019-07-27 (private version)

Added

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

Fixed

  • proper capitalization and spacing when mixin with html bold or italic

Changed

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

[0.18.1] - 2019-07-12 (private version)

Added

  • 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

Changed

  • renamed for freenlg to rosaenlg

[0.18.0] - 2019-07-12 (private version)

Added

  • Indefinite articles in German simplified syntax <…​> parsing

  • Italian verbs

Fixed

  • 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

[0.17.0] - 2019-06-29 (private version)

Added

  • first support of Italian (no verbs)

  • basic support on any language

  • INDEFINITE determiner type in German

Changed

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

[0.16.0] - 2019-05-11 (private version)

Fixed

  • 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

  • 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].

French

  • Cardinal numbers in letters (5 → cinq etc.) support thanks to written-number lib

  • Numbers formatting:

    • Basic support for French ordinal numbers: +value(1, {'ORDINAL_NUMBER':true }) = 1er thanks to numeral lib

    • Support for French cardinal numbers up to 100: +value(21, {'ORDINAL_TEXTUAL':true }) = vingt et unième

  • Improved French verbs support:

    • Any verb of all 3 verb groups are available thanks to lefff derived resource.

    • Supported tenses: PRESENT FUTUR IMPARFAIT PASSE_SIMPLE CONDITIONNEL_PRESENT IMPERATIF_PRESENT SUBJONCTIF_PRESENT SUBJONCTIF_IMPARFAIT PASSE_COMPOSE PLUS_QUE_PARFAIT.

    • For PASSE_COMPOSE and PLUS_QUE_PARFAIT: use aux property (ETRE or AVOIR) and agree property: elles #[+verb(getAnonFP(), {verb: 'sortir', tense:'PASSE_COMPOSE', aux:'ETRE', agree: getAnonFP()})] generates elles sont sorties. If aux is not set, some rules will apply (transitive verbs rather take AVOIR, etc.).

    • Use pronominal:true for pronominal form.

  • French gender shortcuts:

    • setRefGender(PRODUCT, 'bague'); will look for bague in the dictionnary and set PRODUCT gender to F.

    • #[+value('bague', {represents: PRODUIT})] will output bague and set the gender of PRODUIT to F via the dictionnary.

  • French contractions:

    • now manage ce/cet: ce arbre becomes cet arbre

    • manages "h aspiré": le hérisson vs l’hebdomadaire

  • French adjectives:

    • adjPos in value to set the position of the adjective

    • manages "h aspiré":

      • #[+value('homme', {det:'INDEFINITE', adj:'vieux', adjPos:'BEFORE'})] outputs un vieil homme

      • #[+value('hollandais', {det:'INDEFINITE', adj:'vieux', adjPos:'BEFORE'})] outputs un vieil hollandais

  • Very simple integrated POS tagger-like to simplify syntax: [+value('<un vieil hollandais>')] (or [+value('<une vieux hollandais>')]) outputs un vieil hollandais. See .

German

  • German is de_DE.

  • Ponctuation (like English).

  • Dates and numbers.

  • Gender of words (M/F/N) thanks to german-pos-dict.

  • Cases thanks to german-pos-dict.

  • Determiners.

  • getMFN helper (makes the same job than getMorF).

  • Possessives (die Farbe der Gurke).

  • Adjectives (der alten Gurke).

  • Very simple integrated POS tagger-like to simplify syntax: #[+value("<der alte Gurke>", {case:'GENITIVE'})] outputs der alten Gurke. See .

  • Verbs thanks to german-pos-dict - all tenses are supported

English

  • determiners and possessives (thirdPossession)

Changed

  • Some speed optimizations (which are not significant)

  • Refactoring: switched to TypeScript for most of the code

  • Some renaming: please impact your templates

    • shuffle renamed to mix

    • REPRESENTANT: 'ana' renamed to REPRESENTANT: 'refexpr'

  • 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

[0.15.6] - 2018-04-21

Fixed

  • no changes - just fixin Lerna

[0.15.5] - 2018-04-21

Added

  • performance documentation

Fixed

  • 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.

[0.15.4] - 2018-04-20

Fixed

  • 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.

Changed

  • mostly refactoring

  • syno_sentences mixin deprecated, use synz > syn structure

  • assemble mixin deprecated, use itemz > item structure

[0.15.3] - 2018-04-17

Changed

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

  • noFilter parameter is deprecated

[0.15.2] - 2018-04-17

Added

  • French support for months

Changed

  • refactoring, no feature change

[0.15.1] - 2018-04-15

Fixed

  • pug render did not work - fixed.

  • doc is now also generated in one single large PDF file

[0.15.0] - 2018-04-14

Added

  • 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)

    • this is useful while developping

    • force is not compatible with {mode:'sequence'}

    • if the forced alternative is empty, it will still not trigger it

Changed

  • 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

[0.14.1] - 2018-04-12

Fixed

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

  • spaces are automatically added before and after various mixins

[0.14.0] - 2018-04-09

Changed

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

titlecase
  | what is this thing?

[0.13.0] - 2018-04-09

Added

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

eachz elt in elts with { separator: ',', last_separator: 'and' }
  | #{elt}

[0.12.0] - 2018-04-09

Changed

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

[0.11.0] - 2018-04-09

Added

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

  l
    protect bla.bla

[0.10.8] - 2018-04-08

  • just testing the release process - nothing changed

[0.10.7] - 2018-04-08

  • just testing the release process - nothing changed

[0.10.6] - 2018-04-08

  • just testing the release process - nothing changed

[0.10.5] - 2018-04-08

  • just testing the release process - nothing changed

[0.10.4] - 2018-04-05

Fixed

  • path issues to get RosaeNLG’s mixins

[0.10.3] - 2018-04-05

Fixed

  • align versions between rosaenlg and rosaenlg-core

[0.10.2] - 2018-04-05

Fixed

  • wrong French resources path

[0.10.1] - 2018-04-05

Fixed

  • wrong path in include node_modules/rosaenlg/main.pug

[0.10.0] - 2018-04-05

RosaeNLG is now a fork of Pug and not only a Pug add-on. Usage is easier, but maintenance is not.

Changed

  • easier integration:

    • include lib via const rosaenlgPug = require('rosaenlg');

    • then any pug method can be called, for instance rosaenlgPug.renderFile

    • no more (visible) NlgLib object

    • in your template, use include node_modules/rosaenlg/main.pug

  • automatic filtering unless noFilter is set to true

Added

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

  • new structure keywords:

    • synz / syn to define synonyms: replaces +assemble

    • itemz / item to define list of items: replaces +syno_sentences

Fixed

  • a could be transformed into the by compromise in some situations

New syntax example: assemble
- setSize('listElts', 2)
mixin listElts(pos)
  case pos
    when 1
      | A
    when 2
      | B

t #[+assemble('listElts', {separator: ', ', last_separator: 'and'})]

is now:

t
  itemz {separator: ', ', last_separator: 'and'}
    item
      | A
    item
      | B
New syntax example: synonyms
- setSize('synos', 2)
mixin synos(rnd)
  case rnd
    when 1
      | A
    when 2
      | B

t #[+syno_sentences('synos')]

is now:

t
  synz
    syn
      | A
    syn
      | B

[0.9.4] - 2018-03-30

Added

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

[0.9.3] - 2018-03-29

Fixed

  • a/an specific cases, when a is:

    • is at the beginning of a sentence

    • is followed by capitalized text (an AI company)

    • is followed by a protected text

[0.9.2] - 2018-03-29

Added

  • …​ 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.

[0.9.1] - 2018-03-29

Fixed

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

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

[0.9.0] - 2018-03-27

Changed

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

    • mode: 'paragraphs' to generate paragraphs

    • mode: 'sentences' to generate sentences without a new paragraph each time

Added

  • in assemblies, when mode is sentences or paragraphs:

    • new parameters: begin_last_1 and begin_last

    • begin_with_general can now be an array

[0.8.1] - 2018-03-26

Added

  • a first version of the tutorial Tutorial

Fixed

  • 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

[0.8.0] - 2018-03-22

Added

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

[0.7.0] - 2018-03-22

Added

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

  • plural of words in French: œilyeux 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

[0.6.2] - 2018-03-21

Fixed

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

[0.6.1] - 2018-03-20

Added

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

Fixed

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

[0.6.0] - 2018-03-20

Added

  • a/an in English: a industryan industry using compromise lib

[0.5.1] - 2018-03-20

Changed

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

Fixed

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

[0.5.0] - 2018-03-19

Added

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

    • plural of nouns: cranberrycranberries

    • pretty print of numbers: 562,407

    • textual numbers: five thousand five hundred

    • textual and numbered ordinals: 21st, twentieth

    • verbs conjugation with PRESENT PAST FUTURE tenses

  • direct access to compromise lib via util.compromise

  • empty prediction mechanism improvements:

    • better empty prediction mechanism that stops faster using exceptions. PS: it should have improded performance, but it didn’t.

    • user function isNotEmpty() to tell a specific structure is not empty.

    • new mode for synonyms: instead of choosing them randomly, you can trigger then in sequence (first then second etc.). Use setSynoType('syno mixin name', 'sequence') to do that. See

Changed

  • randomSeed param of NlgLib constructor renamed to forceRandomSeed

[0.4.0] - 2018-03-15

Added

  • list of words with their gender in French

  • french adjectives agreement rules (well, most of it)

[0.3.1] - 2018-03-13

Fixed

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

[0.3] - 2018-03-09

Added

  • notion of language: fr_FR or en_US

[0.2] - 2018-03-07

  • initial version