arXiv research summaries for Computation and Language from October 24, 2023. You can find summaries and links to each article here.
Today's research themes - LLM generated - fair warning! 😁
Machine Translation and Speech Recognition
- New methods were proposed for machine translation using large language models, including techniques for controlling gender bias and translating code between programming languages. Advances were made in speech recognition for accented speech.
Language Model Capabilities and Limitations
- Several studies analyzed the capabilities and biases of large language models. Models exhibit social identity biases and struggle with negation and generalization. Emotion recognition was found to help detect synthetic text. Other work looked at how models handle non-natural language input.
Knowledge and Reasoning
- New datasets and methods were introduced to evaluate knowledge graph completion, information extraction, question answering, logical reasoning, and conversational reasoning. Findings show current models still have limitations in robust reasoning.
Model Training Techniques
- Papers explored prompt engineering, contrastive learning, and other techniques to improve model training. Analysis provided insights into how models learn from demonstrations.
Healthcare and Legal Applications
- Datasets and models were developed for generating medical conversations and simplifying legal documents. These have potential to assist doctors and increase access to legal information.
