Phuong Le-Hong

Societal Impact Statement: Promoting agrobiodiversity is a promising strategy for mitigating the negative effects of climate change on global food security. We highlight the central role evolutionary processes play in harnessing the potential of local crops by integrating genomics, archaeology, ethnobotany and traditional ecological knowledge (TEK). Our aim is to empower Indigenous peoples and local communities as key actors in agrobiodiversity and crop innovations, fostering ethical and active collaborations. Local contributions to diversifying the global food system will enhance climate resilience and ensure that the benefits of agricultural innovation are shared more broadly and justly across peoples. Summary: Strengthening diverse agricultural systems can mitigate the effects of accelerated climate change on crop production. Unlocking local crops' potential requires understanding their origins, domestication, fundamental attributes and uses. Here, we review the role of genomics, archaeology and ethnobotany that increased our understanding of crop evolution. We highlight biases in data availability towards globally commercial crops and chart pathways to develop baseline knowledge for local crops. We propose integrating these fields with traditional knowledge to illuminate how cultivation and uses have shaped, and will continue to shape, local crops and promote ethical collaborations. The lessons learnt from the past can help us chart the future of sustainable and climate‐resilient agriculture. Supported with evidence from recent research and traditional knowledge, we propose to orient seed conservation, breeding and accelerated domestication around Indigenous peoples and local communities.

Dependency parsing is essential for language modeling because it offers a structured understanding of the syntactic relationships between words in a sentence. While recent advancements in large language models have greatly advanced the field of natural language processing, dependency parsing remains highly relevant for several key reasons. This paper introduces an effective method for improving dependency parsing which is based on a semantics-aware token embedding model. We propose to incorporate the ConceptNet embeddings which are trained by a retrofitting algorithm into a bidirectional recurrent neural network. The new model outperforms a strong baseline that employs a state-of-the-art method across three dependency treebanks, covering both low-resource and high-resource languages—Indonesian, Vietnamese, and English—achieving an improvement of approximately 1.21% in labeled attachment score.We also show that this method outperforms the popular transformer-based BERT model in capturing syntactic dependency between tokens. The new parser together with all trained models are made available under an open-source license, facilitating community engagement and advancement of natural language processing research for two low-resource languages with around 300 million users worldwide.