The BLEU score (Bilingual Evaluation Understudy) is a metric developed at IBM in 2002 to evaluate machine translations automatically. It measures how many word sequences in a translation also appear in one or more human reference translations and gives a value between 0 and 1 (often shown as 0–100). BLEU is fast and comparable but captures neither meaning nor style well – good paraphrases can score poorly.
ROUGE (Recall-Oriented Understudy for Gisting Evaluation) is a family of metrics for comparing automatically generated texts – especially summaries – with reference texts written by people. It measures how many words or word sequences overlap, for example single words in ROUGE-1 or the longest common subsequence in ROUGE-L. The score is quick to compute, but it does not reliably capture paraphrases with the same meaning or factual accuracy.
In AI, benchmarking means systematically comparing models or systems using defined tasks and criteria. Public benchmarks measure things like general knowledge, reasoning or coding ability; for practical use, however, test cases from your own day-to-day work matter most. Typical criteria are accuracy, completeness, response time and cost. This helps decide whether a smaller, cheaper model is good enough for a task.
The F-score is a metric for evaluating classification models. The common F1 score is the harmonic mean of precision (how many of the predicted hits are correct) and recall (how many of the actual hits were found), and ranges from 0 to 1. It is especially useful with imbalanced data: a spam filter that almost never flags spam can have high accuracy but a low F1 score.
Natural language processing is the branch of AI that enables computers to analyse, understand and generate human language in text or speech. It covers tasks such as translation, classification, summarisation, sentiment analysis and question answering. Rules and statistical methods once dominated; today the field is shaped by large language models.
