ConLoan-LV: A Contrastive Dataset for Latvian Language Loanwords, Code-switching, and Named Entities

Replicating and extending the ConLoan methodology, the dataset contains 353 manually validated sentences in the baseline version and 676 in the extended version, with all sentences sourced from the LVK2022 corpus. Each entry is enriched with labels for material borrowings (LOAN), while the extended version adds labels for code-switching (CS) and named entities (NE).

Furthermore, the dataset includes native-language semantic equivalents for loanwords and English translations, providing a parallel structure for comparative analysis. This resource is intended for training and benchmarking language models in identifying non-native lexical elements within Latvian language texts.