TMX and TBX

A translation memory leaves and arrives as TMX, a glossary as TBX: the exchange formats translation tools read and write. Both readers refuse a file that declares a document type before parsing it, fetch nothing from the network and expand no entity.

TMX 1.4b

TmxWriter writes the units of one source language — units of another source language are left out:

use Phil\TranslationMemory\LanguageTag;
use Phil\TranslationMemory\Tmx\TmxReader;
use Phil\TranslationMemory\Tmx\TmxWriter;

$xml = new TmxWriter()->write(LanguageTag::of('en'), $units);   // iterable<MemoryUnit>
$units = new TmxReader()->read($xml);                          // list<MemoryUnit>

<tmx version="1.4">
  <header creationtool="phil/translation-memory" creationtoolversion="0.1" segtype="sentence" o-tmf="phil" adminlang="en" srclang="en" datatype="html"/>
  <body>
    <tu>
      <prop type="x-origin">page:1</prop>
      <tuv xml:lang="en"><seg>A <bpt i="1">&lt;b&gt;</bpt>bold<ept i="1">&lt;/b&gt;</ept> move</seg></tuv>
      <tuv xml:lang="fr"><seg>Un coup <bpt i="1">&lt;b&gt;</bpt>audacieux<ept i="1">&lt;/b&gt;</ept></seg></tuv>
    </tu>
  </body>
</tmx>

Each unit is a <tu> with its source and its target as <tuv>; its origin, when known, a x-origin property. Tags stay out of the words a tool may edit: an opening tag and the closing tag that matches it become <bpt> and <ept> sharing a number, any other tag a <ph>, each holding the HTML it stands for.

TmxReader reads TMX 1.4b, and the earlier versions as far as they agree. The source of a <tu> is the <tuv> of its own srclang, or of the header's; with *all*, its first <tuv>. Every other <tuv> gives a unit from that source into its language, xml:lang or the older lang. Inline codes come back as the HTML they hold; <hi> and <sub> keep their words. A <tuv> without words, or a <tu> without its source, is left out. A file holds 100 000 units at most (TmxReader::MAX_UNITS).

To load a file into a memory:

foreach (new TmxReader()->read($xml) as $unit) {
    $store->add($unit);
}

InvalidTmx is thrown for a text that is not well-formed XML with a <tmx> root, a document type, a missing <header> or <body>, a language that is not a tag, or too many units.

TBX

TbxWriter writes a glossary as TBX of ISO 30042:2019, in the TBX-Basic dialect and its DCA style; TbxReader reads the terms of one pair of languages from that TBX or from the TBX of 2008, whose root is <martif>.

use Phil\TranslationMemory\Tbx\TbxReader;
use Phil\TranslationMemory\Tbx\TbxWriter;

$xml = new TbxWriter()->write($glossary);
$glossary = new TbxReader()->read($xml, LanguageTag::of('en'), LanguageTag::of('fr-CA'));

Each term is a <termSec> (<tig> or <ntig> in 2008) with its administrativeStatus. When reading, the source term is the preferred one, or else the first that is not deprecated; the other source terms that are not deprecated are its forms. The target terms that are not deprecated are its translations; the deprecated, superseded and not recommended ones are refused — a term whose every target term is refused reads as kept. An entry without both languages is left out. A language matches its parent: an entry with a fr section and no fr-CA one is read for fr-CA.

TBX has no place for case: a case-sensitive term is read back as one that ignores case. A file holds 100 000 entries at most (TbxReader::MAX_ENTRIES). InvalidTbx is thrown for a text that is not well-formed XML with a <tbx> or <martif> root, a document type, a language that is not a tag, or too many entries.