Science News
Multilingual and Cross-Lingual Citation Needed Detection on Wikipedia for Lower-Resource Languages: experiments showed that SLMs fine-tuned with an encoder-style objective substantially outperform prompted LLMs across languages Read Full Article #MultilingualNLP…
⭐ A new multilingual corpus, MCN, has been released, covering 18 languages for "Citation Needed Detection" (CND) on Wikipedia, specifically addressing lower-resource languages.
⭐ Research shows that small language models (SLMs) fine-tuned for CND significantly outperform large language models (LLMs) across multiple languages.
⭐ The study also demonstrates that SLMs trained primarily on English can achieve strong performance in cross-lingual CND, even with minimal target-language data.
#AI #NLP #Wikipedia
⭐ Research shows that small language models (SLMs) fine-tuned for CND significantly outperform large language models (LLMs) across multiple languages.
⭐ The study also demonstrates that SLMs trained primarily on English can achieve strong performance in cross-lingual CND, even with minimal target-language data.
#AI #NLP #Wikipedia
Science News
New method TOHA detects LLM hallucinations in RAG systems by analyzing attention map topology (MTop-Div). It requires no extra training, matches heavier computation methods, and is open-sourced. Read Full Article #LLMHallucinationDetection #RAGSystems …
🧠 Researchers led by Alexandra Bazarova have introduced TOHA, a new method for detecting "hallucinations" (outputs not supported by context) in large language models (LLMs), especially in Retrieval-Augmented Generation (RAG) settings.
📊 TOHA uses a "topological divergence metric" on attention graphs to analyze the structural properties of LLM attention patterns. Higher divergence in specific attention areas correlates with unfaithful outputs, regardless of the dataset.
🏆 Experiments on question answering and summarization tasks show TOHA achieves state-of-the-art or competitive results, requiring minimal annotated data and computational resources.
#LLM #AI #NLP
📊 TOHA uses a "topological divergence metric" on attention graphs to analyze the structural properties of LLM attention patterns. Higher divergence in specific attention areas correlates with unfaithful outputs, regardless of the dataset.
🏆 Experiments on question answering and summarization tasks show TOHA achieves state-of-the-art or competitive results, requiring minimal annotated data and computational resources.
#LLM #AI #NLP