Search papers, labs, and topics across Lattice.
This study investigates the transition from the tripartite grammatical gender system of Latin to the bipartite system in Romance languages using an interpretable deep learning framework. The authors identify that conventional tokenization methods are inadequate for this historical analysis and introduce a novel tokenizer that enhances performance. By analyzing morphological features and part-of-speech contributions, the research delineates how gender information is distributed between lemmas and their contexts, providing insights into linguistic evolution.
Conventional tokenization fails in low-resource historical settings, but a new approach reveals deeper insights into gender evolution in Romance languages.
The diachronic evolution from Latin to the Romance languages involved a restructuring of the grammatical gender system from a tripartite configuration (masculine, feminine, neuter) to a bipartite one (masculine, feminine) in most Romance languages. In this work, we introduce an interpretable deep learning framework to investigate this phenomenon at both lexical and contextual levels. First, we show that conventional tokenization strategies are insufficiently robust for this low-resource historical setting, and that our proposed tokenizer improves performance over these baselines. At the lexical level, we evaluate the contribution of morphological features to gender prediction. At the contextual level, we quantify the contributions of different part-of-speech categories to grammatical gender prediction. Together, these analyses characterize the distribution of gender information between the lemma and its sentential context. We make our codebase, datasets, and results publicly available at \href{https://github.com/ahan-2000/Lost-in-Translation-}{https://github.com/ahan-2000/Lost-in-Translation-}.
