Dense vector representations for textual data are crucial in modern NLP. Word embeddings and sentence embeddings estimated from raw texts are key in achieving state-of-the-art resultsin various tasks requiring semantic understanding. However, obtaining embeddings at the document level is challenging due to computational requirements and lack of appropriate data. Instead, most approaches fall back on computing document embeddings based on sentence representations. Although there exist architectures and models to encode documents fully, they are in general limited to English and few other high-resourced languages. In this work, we provide a systematic comparison of methods to produce document-level representations from sentences based on LASER, LaBSE, and Sentence BERT pre-trained multilingual models. We compare input token number truncation, sentence averaging as well as some simple windowing and in some cases new augmented and learnable approaches, on 3 multi- and cross-lingual tasks in 8 languages belonging to 3 different language families. Our task-based extrinsic evaluations show that, independently of the language, a clever combination of sentence embeddings is usually better than encoding the full document as a single unit, even when this is possible. We demonstrate that while a simple sentence average results in a strong baseline for classification tasks, more complex combinations are necessary for semantic tasks

Multilingual language models have shown impressive cross-lingual transfer ability across a diverse set of languages and tasks. To improve the cross-lingual ability of these models, some strategies include transliteration and finer-grained segmentation into characters as opposed to subwords. In this work, we investigate lexical sharing in multilingual machine translation (MT) from Hindi, Gujarati, Nepali into English. We explore the trade-offs that exist in translation performance between data sampling and vocabulary size, and we explore whether transliteration is useful in encouraging cross-script generalisation. We also verify how the different settings generalise to unseen languages (Marathi and Bengali). We find that transliteration does not give pronounced improvements and our analysis suggests that our multilingual MT models trained on original scripts are already robust to cross-script differences even for relatively low-resource languages.