AI is fast becoming the go-to tool for English educators who want accurate assessment tools when measuring the communicative ability of students, but this is not without a catch.

AI platforms are as good as the quality of data they are trained with, which can make them reliable and insightful evaluations of a student’s English language proficiency.

What sets AI testing platforms like Envoy, a new online English language test, apart is its ability to use open-ended responses rather than constructed responses. While constructed responses are useful, they don’t allow for the construct to be adequately tested.

When assessing communicative ability of students, it is important to test language as it is used in communication in situations similar to real life where people are required to speak or write at lengths more than single phrases or sentences. In other words to use language for communication students are required to create open-ended and extended discourse, for example give presentations, talk about a topic or write a response in the form of an email, a paragraph or an essay. Envoy’s algorithms have been trained to assess open-ended responses.

This gives Envoy an advantage over many other AI English language proficiency tests when it comes to assessing productive skills, said Reza Tasviri, IDP’s Head of Assessment.

“Many of AI English tests only measure constructed responses, such as read-alouds and cloze tests or gap fills which are useful to an extent, but may only provide a partial picture of language ability” he explained.

Envoy also uses analytic criterion-referenced scoring when assessing communicative ability of students. What’s more, these methods of scoring provide true and interpretable reflections of how a student is learning and provide scores that serve diagnostic needs and inform remedial next steps for educators.

This makes Envoy’s AI not only reliable, but fair. With input from a diverse range of students, the AI has been trained on a variety of first languages, accents, ages and other demographics. This means Envoy better reflects the diversity of students who use its AI speaking test while also removing bias.

Diversity can also be seen in its scoring structures with Envoy’s human raters all hailing from a range of backgrounds. Additionally, independent raters who score training data do so without any demographic data and Envoy’s algorithms are blind to demographic data during the scoring process.

Anti-cheating measures are integrated throughout the program so educators can be assured test results are genuine. Envoy asks open-ended questions to make it harder for students to cheat on the tests and natural speech training means students trying to beat algorithms with multiple repetitions, loud or fast speech, or pre-prepared answers will not succeed.

This is complemented by Envoy’s ability to help educators save time on administrative tasks and marking. As opposed to traditional testing, Envoy’s AI technology can be used to quickly and efficiently score tests with an accurate and detailed score report generated within two hours of test completion.

It’s a win-win for educators and students alike, with more value time given back to the class.

Most importantly, Envoy’s AI abilities are constantly growing. New data is used to train the AI and refine algorithms so Envoy can evolve and continue to support busy learning environments.

Contact us to connect with a team member on how Envoy can help provide a fair and reliable assessment of your student's English proficiency.