Cross-lingual transfer of abstractive summarizer to less-resource language

Automatic text summarization extracts important information from texts and presents the in form of a summary. Abstractive approaches progressed significantly by switching to deep neural networks, but results are not yet satisfactory, especially for languages where large training sets do exist. In several natural language processing tasks, cross-lingual model transfer is successfully applied less-resource languages. For summarization, was attempted due non-reusable decoder side models that cannot correct target generation. our work, we use pre-trained English based on networks sequence-to-sequence architecture summarize Slovene news articles. We address problem inadequate using an additional evaluation generated language. test with different amounts data fine-tuning. assess automatic measures conduct small-scale human evaluation. shows summaries best useful quality similar trained only Human generates high accuracy acceptable readability. However, other abstractive models, perfect may occasionally produce misleading or absurd content.