Uploads training files to customize a translation model. You can customize a model with a forced glossary or with a parallel corpus:

Use a forced glossary to force certain terms and phrases to be translated in a specific way. You can upload only a single forced glossary file for a model. The size of a forced glossary file for a custom model is limited to 10 MB.
Use a parallel corpus when you want your custom model to learn from general translation patterns in parallel sentences in your samples. What your model learns from a parallel corpus can improve translation results for input text that the model has not been trained on. You can upload multiple parallel corpora files with a request. To successfully train with parallel corpora, the corpora files must contain a cumulative total of at least 5000 parallel sentences. The cumulative size of all uploaded corpus files for a custom model is limited to 250 MB.

Depending on the type of customization and the size of the uploaded files, training time can range from minutes for a glossary to several hours for a large parallel corpus. To create a model that is customized with a parallel corpus and a forced glossary, customize the model with a parallel corpus first and then customize the resulting model with a forced glossary.

You can create a maximum of 10 custom models per language pair. For more information about customizing a translation model, including the formatting and character restrictions for data files, see Customizing your model.

Supported file formats

You can provide your training data for customization in the following document formats:

TMX (.tmx) - Translation Memory eXchange (TMX) is an XML specification for the exchange of translation memories.
XLIFF (.xliff) - XML Localization Interchange File Format (XLIFF) is an XML specification for the exchange of translation memories.
CSV (.csv) - Comma-separated values (CSV) file with two columns for aligned sentences and phrases. The first row must have two language codes. The first column is for the source language code, and the second column is for the target language code.
TSV (.tsv or .tab) - Tab-separated values (TSV) file with two columns for aligned sentences and phrases. The first row must have two language codes. The first column is for the source language code, and the second column is for the target language code.
JSON (.json) - Custom JSON format for specifying aligned sentences and phrases.
Microsoft Excel (.xls or .xlsx) - Excel file with the first two columns for aligned sentences and phrases. The first row contains the language code.

You must encode all text data in UTF-8 format. For more information, see Supported document formats for training data.

Specifying file formats

You can indicate the format of a file by including the file extension with the file name. Use the file extensions shown in Supported file formats.

Alternatively, you can omit the file extension and specify one of the following content-type specifications for the file:

TMX - application/x-tmx+xml
XLIFF - application/xliff+xml
CSV - text/csv
TSV - text/tab-separated-values
JSON - application/json
Microsoft Excel - application/vnd.openxmlformats-officedocument.spreadsheetml.sheet

For example, with curl, use the following content-type specification to indicate the format of a CSV file named glossary:

--form "forced_glossary=@glossary;type=text/csv".

Language Translator | IBM Cloud API Docs