🤖 AI Summary
To address the challenges of ambiguous intent definitions, high annotation costs, and data scarcity in low-resource language intent classification, this paper proposes a zero-shot retrieval-based intent identification method. Instead of relying on explicit intent labels, the approach models each intent as a set of historically annotated queries and performs intent inference via dense retrieval—matching an input query to its most semantically similar labeled queries in a shared latent space. The method employs a multilingual semantic encoder to produce cross-lingually aligned query embeddings, enhances semantic consistency through contrastive learning, and accelerates retrieval using approximate nearest neighbor (ANN) search. Evaluated on eight low-resource languages, it achieves an average F1 score of 72.4%, outperforming zero-shot fine-tuning baselines by 18.6 percentage points; remarkably, it attains near fully supervised performance using only 10% of the labeled data. This work is the first to reformulate intent classification as an annotation-free, cross-lingual similarity search task.
📝 Abstract
Intent classification is an important component of a functional Information Retrieval ecosystem. Many current approaches to intent classification, typically framed as a classification problem, can be problematic as intents are often hard to define and thus data can be difficult and expensive to annotate. The problem is exacerbated when we need to extend the intent classification system to support multiple and in particular low-resource languages. To address this, we propose casting intent classification as a query similarity search problem - we use previous example queries to define an intent, and a query similarity method to classify an incoming query based on the labels of its most similar queries in latent space. With the proposed approach, we are able to achieve reasonable intent classification performance for queries in low-resource languages in a zero-shot setting.