--- tags: - sentence-transformers - sentence-similarity - feature-extraction - cross-lingual - historical-nlp - knowledge-distillation - dataset_size:480534 - loss:MSELoss language: - grc - la - en base_model: bowphs/SPhilBerta widget: - source_sentence: ἐκνεύει πάλιν. sentences: - >- I found out that he was accused concerning questions of their law, but had nothing charged against him deserving of death or chains. - he is drawing back. - >- In this book I have fully set forth the mechanical methods which I could furnish, and which I thought most useful in times of peace and war. - source_sentence: καὶ εἶπεν ὁ βασιλεύς Ἐπίστρεψον, στηλώθητι ὧδε· καὶ ἐπεστράφη καὶ ἔστη. sentences: - >- Indeed, he often lapses unawares into the manner of Xenarchus, as, for instance, when he says he thinks it was a bad omen for the Athenians that Nicias, whose name was derived from victory, declined at first to head their expedition also that, by the mutilation of the Hermae, Heaven indicated to them in advance that by the hands of Hermocrates the son of Hermon they were to suffer most of their reverses during the war - >- They pay but moderate duties both on the imports and exports from Keltica which are ivory bracelets and necklaces, amber, vessels of glass, and small wares - And the king said, “Turn aside; stand here.” And he turned and stood there. - source_sentence: >- Quo maerore perculsus, velut primo adventantium malorum auspicio, murmurans querulis vocibus saepe audiebatur: nihil se plus assecutum, quam ut occupatior interiret. sentences: - >- Overwhelmed by sorrow at this, the first omen, as it were, of approaching ills, he was often heard to mutter in complaining tones that he had gained nothing, except to die with heavier work. - >- for never again shall I bury one whom I love more, no! nor one more loyal to me - >- And in truth we could do nothing more pious than to give attention to this matter, nothing more impious than to disregard it. - source_sentence: >- μαθὼν δὲ ἐκ τῆς μαντικῆς τὸν κεκλοφότα πρὸς Μαῖαν εἰς Κυλλήνην παραγίνεται, καὶ τὸν Ἑρμῆν ᾐτιᾶτο. sentences: - >- As it is, thinking ourselves despised, we not only treat harshly wife and slaves and friends, but also through rage often fall out with innkeepers and sailors and drunken muleteers we even rage against dogs that bark at us and asses that jostle us, like the man who wished to beat the assdriver, but when the driver cried out, I am an Athenian, indicated the ass and said, You at any rate are not an Athenian, and fell to beating it with many blows. - >- Three walls were easily built, because so many hands were continually at work on them, for it was summer time and there was nothing, either the weather or any lack of workers, to hinder the building. Then bringing up their machinery they shook the walls of the city and tried in every way to take it. - >- Having discovered the thief by divination, Apollo came to Maia at Cyllene and accused Hermes. - source_sentence: >- Νομίζω οὖν τοῦτο καλὸν ὑπάρχειν διὰ τὴν ἐνεστῶσαν ἀνάγκην, ὅτι καλὸν ἀνθρώπῳ τὸ οὕτως εἶναι. sentences: - >- and the daffodil, νάρκισσος, because it benumbs the nerves and causes a stupid narcotic heaviness in the limbs - He is thus seen to be six years the junior of Isocrates. - >- I suppose therefore that this is good because of the present distress -- that it is good for a man to remain as he is: pipeline_tag: sentence-similarity library_name: sentence-transformers model-index: - name: SentenceTransformer based on bowphs/SPhilBerta results: - task: type: knowledge-distillation name: Knowledge Distillation dataset: name: Latin to Ancient Greek type: la-grc metrics: - type: negative_mse value: -0.13404905796051025 name: Negative Mse --- # DistilSPhilBERTa Latin and Ancient Greek cross-lingual sentence mining and similarity detection is a relatively new field. In our paper [Evaluating Latin and Ancient Greek Sentence Alignment through Parallel Sentence Mining](https://aclanthology.org/2026.nlp4dh-1.11/) we introduce DistilSPhilBERTa, a [sentence-transformers](https://www.SBERT.net) model finetuned from [bowphs/SPhilBerta](https://huggingface.co/bowphs/SPhilBerta) for exactly these tasks. It achieves significantly better results in our custom [sentence mining benchmark](https://github.com/TUM-NLP/latin-greek-mining) than SPhilBERTa and other classical cross-lingual language models. It maps sentences & paragraphs to a vector space and can be used for semantic textual similarity, semantic search, parallel sentence mining and more. ## Model Details ### Model Description - **Model Type:** Sentence Transformer - **Base model:** [bowphs/SPhilBerta](https://huggingface.co/bowphs/SPhilBerta) - **Maximum Sequence Length:** 128 tokens - **Output Dimensionality:** 768 dimensions - **Similarity Function:** Cosine Similarity - **Training Dataset:** [grosenthal/latin_english_parallel](https://huggingface.co/datasets/grosenthal/latin_english_parallel), [English-Greek Dataset](https://github.com/kevinkrahn/ancient-greek-datasets) ## Usage ### Direct Usage (Sentence Transformers) With the Sentence Transformers library installed, you can use the model in the following way: ```python from sentence_transformers import SentenceTransformer model = SentenceTransformer("sebastian-reichbauer/DistilSPhilBERTa") # Run inference sentences = [ 'Νομίζω οὖν τοῦτο καλὸν ὑπάρχειν διὰ τὴν ἐνεστῶσαν ἀνάγκην, ὅτι καλὸν ἀνθρώπῳ τὸ οὕτως εἶναι.' ] embeddings = model.encode(sentences) print(embeddings) ```