The article discusses the development of Natural Language Generation (NLG) systems, which produce texts in English or other languages from computer-accessible data. These systems are commonly used to assist human authors in creating routine documents, such as business letters and weather reports, and as interactive tools to explain information to non-expert users in fields like software engineering and medicine. From a technical standpoint, NLG systems typically perform three main tasks: content determination and text planning, sentence planning, and realization. Content determination and text planning involve deciding what information to communicate and how to structure it rhetorically. This can be done through simple hard-coded planners, more sophisticated rule-based systems, or intermediate schema-based languages. Sentence planning focuses on making the text fluent by using conjunctions, pronouns, and discourse markers. Realization involves generating grammatically correct sentences, considering aspects like punctuation, morphology, agreement, and reflexives. The article highlights that various techniques exist for each NLG task, ranging from simple to highly sophisticated. The choice of technique depends on the application's requirements, such as the need for fluency, information filtering, and syntactic variety. A skilled NLG engineer selects the most appropriate methods based on the application's needs and available resources. The paper concludes that no single approach is universally better, and the effectiveness of NLG systems depends on the specific context and requirements of the task.The article discusses the development of Natural Language Generation (NLG) systems, which produce texts in English or other languages from computer-accessible data. These systems are commonly used to assist human authors in creating routine documents, such as business letters and weather reports, and as interactive tools to explain information to non-expert users in fields like software engineering and medicine. From a technical standpoint, NLG systems typically perform three main tasks: content determination and text planning, sentence planning, and realization. Content determination and text planning involve deciding what information to communicate and how to structure it rhetorically. This can be done through simple hard-coded planners, more sophisticated rule-based systems, or intermediate schema-based languages. Sentence planning focuses on making the text fluent by using conjunctions, pronouns, and discourse markers. Realization involves generating grammatically correct sentences, considering aspects like punctuation, morphology, agreement, and reflexives. The article highlights that various techniques exist for each NLG task, ranging from simple to highly sophisticated. The choice of technique depends on the application's requirements, such as the need for fluency, information filtering, and syntactic variety. A skilled NLG engineer selects the most appropriate methods based on the application's needs and available resources. The paper concludes that no single approach is universally better, and the effectiveness of NLG systems depends on the specific context and requirements of the task.