Manual testing of software requirements written in natural language for agile or any other methodology requires more time and human resources. This leaves the testing process error prone and time consuming. For satisfied end users with bug‐free software delivered on time, there is a need to automate the test oracle process for natural language or informal requirements. The automation of the test oracle is relatively easier with formal requirements, but this task is difficult to achieve with natural language requirements. This study proposes an approach called Restricted Natural Language Agile Requirements Testing (ReNaLART) to automate the test oracle from restricted natural language agile requirements. For this purpose, it uses an existing user story template with some modifications for writing user stories. This helps in identifying test input and expected output for a user story. For comparison of expected and observed outputs it makes use of a regex pattern and string distance functions. It is capable of assigning different types of verdicts automatically depending upon the similarity/dissimilarity between observed and expected outputs of user stories. ReNaLART is validated using several case studies of different domains, namely, OLX Pakistan, Mental Health Tests, McDelivery Pakistan, BlueStacks, Power Searching with Google, TensorFlow Playground, w3Schools 2018 offline and Touch'D. It revealed several faults in five of the above listed eight applications. Plus, the proposed test oracle on an average took 0.02 s for test data generation, expected output generation and verdict assignment. Both these facts show the fault revealing effectiveness and efficiency of ReNaLART.

中文翻译：

---

根据受限的自然语言敏捷性要求自动执行测试预告片

手动测试以自然语言编写的敏捷或任何其他方法的软件需求需要更多的时间和人力资源。这使测试过程容易出错且耗时。对于使用及时交付的无错误软件的满意最终用户而言，有必要针对自然语言或非正式要求自动执行测试Oracle流程。对于正式的需求，测试oracle的自动化相对容易一些，但是使用自然语言的需求则很难实现这一任务。这项研究提出了一种称为“受限自然语言敏捷需求测试”的方法（ReNaLART），以根据受限的自然语言敏捷性要求自动执行测试Oracle。为此，它使用经过修改的现有用户故事模板来编写用户故事。这有助于识别用户故事的测试输入和预期输出。为了比较预期和观察到的输出，它使用了正则表达式模式和字符串距离函数。它能够根据观察到的和预期的用户故事输出之间的相似性/不同性自动分配不同类型的判决。ReNaLART已使用多个不同领域的案例研究进行了验证，分别是OLX巴基斯坦，心理健康测试，巴基斯坦的McDelivery，BlueStacks，Google的Power Searching，TensorFlow Playground，w3Schools 2018离线版和Touch'D。它揭示了上面列出的八个应用程序中的五个中的几个故障。另外，建议的测试预告片平均花费0.02 s来进行测试数据生成，预期输出生成和判决分配。这两个事实都显示了ReNaLART的缺陷揭示了其有效性和效率。