Computational Linguistics

Computational linguistics sits at the intersection of linguistics, computer science and artificial intelligence, aiming to model the structure and use of human language in computational systems. From its origins in rule-based grammars and machine translation experiments in the mid-20th century, the field has evolved through statistical methods—such as n-gram models and hidden Markov models—to today’s deep learning approaches that leverage large-scale neural architectures. Researchers examine the formal properties of syntax and semantics, develop algorithms for parsing and generation, and employ distributional semantics to derive meaning from patterns of word co-occurrence. Applications are far-reaching: they include machine translation, information retrieval, dialogue systems, sentiment and stance analysis, and digital humanities. Current challenges involve modelling context and discourse, handling low-resource languages, ensuring interpretability of neural models and mitigating algorithmic bias. By combining symbolic insights with data-driven learning, computational linguistics continues to enhance global communication technologies—powering voice assistants, automated translation, content moderation and more—while advancing our theoretical understanding of the nature of language.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Researchers applied graph convolutional networks and feature-propagation techniques to detect stances on vaccination in a low-resource dialect during the COVID-19 rollout. By modelling both user interactions and textual cues from retweet networks, the system achieved high accuracy in classifying support, opposition and neutrality, and quantified peaks of polarisation correlating with public health announcements. Another study targeted the “silent majority” of passive social-media users by integrating collaborative filtering with graph convolutional networks. Analysing user-topic relations in datasets from Chilean constitutional referendums, the method improved stance-prediction accuracy by over nine per cent compared with standard baselines, revealing how non-posting users influence perceptions of collective opinion. A further investigation evaluated unfine-tuned large language models for zero-shot stance detection on political statements drawn from Twitter timelines. Leveraging a natural language inference framework without labelled training data, the approach generalised across nine electoral contexts in multiple countries, producing performance on a par with supervised classifiers and demonstrating the potential of zero-shot inference for emerging or under-sampled topics.

Computational Linguistics publication trend

The graph below shows the total number of articles in computational linguistics across all publications each year (not limited to Nature Index journals).

Technical terms

Natural language processing (NLP): Algorithmic methods for analysing, understanding and generating human language data.

Stance detection: The task of classifying text as expressing support, opposition or neutrality towards a specified target.

Graph convolutional network (GCN): A neural architecture that performs convolution operations over graph-structured data to capture relational patterns among entities such as users or words.

Zero-shot classification: A technique enabling models to assign labels to instances unseen during training by leveraging generalised semantic knowledge.

Pre-trained language model: A neural network trained on large corpora to learn general linguistic representations before being adapted to downstream tasks.

References

1. Evaluating large language models for user stance detection on X (Twitter). Machine Learning (2024).
2. Unveiling the silent majority: stance detection and characterization of passive users on social media using collaborative filtering and graph convolutional networks. EPJ Data Science (2024).