Datasets:

---

ayymen

/

Weblate-Translations

Like 24

Hello @ZJaume , I'm glad you found the dataset useful 🙂 if I knew it was being used at Mozilla I would have kept it updated at least on a quarterly basis. Looking now at their API, it seems the dataset has indeed grown by roughly 52%. Hopefully there are also new languages there. I have a couple of questions if you don't mind:

• The first time I uploaded the data I tried to clean up the codes but quickly gave up, there's just so much stuff here - pirate speak, "uwu" speak, emoji language, multiple subsets for the same language, etc... so I'd like to know your thoughts in case you've looked at the different subsets or compared them. For example, are the various suffixes attached to lang codes informative? for instance in the en-de direction there's @ formal, @ informal, _1901, _rude and many others. Your feedback can help me simplify the codes and improve documentation to make the dataset easier to use.
• Are you interested in using other multilingual localization datasets such as this one? there are other data sources as well.

I will let you know once I upload a new version.

Great news then!

Regarding the questions:

• The language codes seem pretty messy, but taking into account that for the majority of the languages, 90% of the data is in the en-xx code, there's not much need in cleaning that up. In the future, we may train more low-resource languages that might raise issues about the language code mess, so I guess I could try to report them to you in the future. But anyway, I made a Huggingface dataset importer for the pipeline that let's me add several subsets of a dataset if needed, so I can, let's say, add en-pt and all other subsets with country BCP code suffixes in the same pipeline configuration without too much trouble.
• Yes, we are interested in more localization datasets. But small datasets like the one you are pointing out are much very important for now (if they have only a few thousand pairs per language, for example 1.3k sentences for Hungarian won't make much difference), until we start training very low resource languages, which may take some time until we start.
• I was also wondering if you used any of the metadata available in Weblate. The exported TMX has a field that tells if the translation has been finished and that could work as double quality check. Although I'm not sure if can be used if decreases each language pair data by 20% or more. I guess it will depend on the quality of discarded sentence pairs, if a lot of bad translations get discarded, it's good. Maybe it could be added as a dataset column.

Indeed, we are interested in any kind of parallel data, not just localization data.

Thanks for your response.

I'm currently finishing processing the dataset and so far I've added the context, translator_comments, and developer_commentscolumns found in the translation files. I'd like to add the field you mentioned but I couldn't find it. Do you have an link to a file that contains it?

Not sure how you are accessing the data, but downloading xliff files from each project, one metadata field in the target is "state".

https://hosted.weblate.org/download/dialect/-/zgh/?format=zip:xliff11

https://hosted.weblate.org/download/f-droid/-/ca/?format=zip:xliff11

        <target state="translated">ⴰⵙⴰⵏⴼ ⵏ ⵓⵙⵓⵖⵍ</target>
        <target state="needs-translation">ⵙⵙⵉⴷⴼ ⵜⴰⵙⴰⵔⵓⵜ ⵏ API ⵉ ⵜⵏⴰⴼⵓⵜ ⵏ ⵓⵙⵓⵖⵍ.</target>
        <target state="needs-translation">ⵙⵙⵏⵖⵍ ⵜⴰⵙⵓⵖⵍⵜ</target>
        <target state="translated">ⴰⵎⵓⵜⵜⵉ</target>
        <target state="needs-translation">ⵙⵙⵏⵖⵍ ⵜⴰⵙⵓⵖⵍⵜ</target>
        <target state="needs-translation">ⵕⵥⵎ ⵉⵙⵎⵢⴰⴼⵓⵜⵏ</target>
        <target state="translated">ⴰⵙⵓⵖⵍ ⵏⴳⵔ ⵜⵓⵜⵍⴰⵢⵉⵏ</target>
        <target state="translated">ⵜⵉⵎⵥⵍⴰⵢ:</target>
        <target state="translated">ⵉⵙⵓⵏⴰⴼ ⵏ ⵜⵏⴰⵙⵜ</target>
        <target state="translated">ⵙⵍⵍ</target>
        <target state="needs-translation">ⴰⵣⴳⴰⵍ ⴳ ⵡⴰⴳⴰⵎ ⵏ ⵜⵏⴰⴼⵓⵜ ⵏ ⵓⵙⵓⵖⵍ</target>

not sure if it's reliable and useful for filtering. I took a look at 4 projects and 4 languages and most of the "need translation" states are already translated, not sure if they are just translated and not marked by the user as "translated" or what. Maybe it could work as a safeguard, I've seen projects in the past that fill up the non-English translations with English data until they are translated. But anyway, it could be nice to have and use it afterwards if needed.

I was actually using csv files, which don't have this field.

I checked the first example you provided and it seems that state is actually useful: the needs-translation state seems to indicate the translation was just a fuzzy match, not a manual translation. This could significantly improve the quality of the data when used as a filter.

There's also an approved field that could be useful. I'll add both.

Hi @ZJaume , just wanted to give you a quick update: I've now downloaded almost half of the raw data. Unfortunately the process is a bit slow because I have to respect Weblate's API rate limits and also because xliff files are much bigger than csv files.

I'm also trying to be more careful with the data preprocessing this time. There are some target strings in the translation files that are not meant to be translations, but rather act as metadata. For example "translator-credits" or "translator_credits".

I have uploaded most of the new data here, 66 configs failed to upload - mostly English, Greek, and Esperanto data. I'm working on fixing the issue.