[Bug]: Language Order in OpusParallelCorpus

[chelseagzr]

Describe the bug

When instantiating an OpusParallelCorpus with language pairs such as ("de", "en") or ("en", "de"), the resulting corpus consistently has German as the first language and English as the second language, regardless of the input order.

Upon reviewing the source code of OpusParallelCorpus, it appears that the following code:

if l1 > l2:
    l1, l2 = l2, l1

forces the languages to be ordered lexicographically based on their language codes. This results in German being treated as the first language in both ("de", "en") and ("en", "de").

To Reproduce

from flair.datasets import OpusParallelCorpus

corpus_de_en = OpusParallelCorpus(
    dataset="tatoeba",
    l1="de",
    l2="en",
    max_tokens_per_doc=512,
)

corpus_en_de = OpusParallelCorpus(
    dataset="tatoeba",
    l1="en",
    l2="de",
    max_tokens_per_doc=512,
)

# Both corpora consist of (German, English) pairs
print(corpus_de_en.train[0])
>> DataPair: 'Sentence[5]: "Ich muss schlafen gehen."' + 'Sentence[7]: "I have to go to sleep."'
print(corpus_en_de.train[0])
>> DataPair: 'Sentence[5]: "Ich muss schlafen gehen."' + 'Sentence[7]: "I have to go to sleep."'

Expected behavior

Sentence pairs in corpus_de_en are (German, English) while sentence pairs in corpus_en_de are (English, German)

Logs and Stack traces

Screenshots

No response

Additional Context

No response

Environment

Versions:

Flair

0.15.1

Pytorch

2.6.0+cu124

Transformers

4.49.0

GPU

False