Automated speech testing is reliable, fast, and cost-effective. Hence why it’s becoming more and more popular. But how do automated English tests work? What’s the underlying technology? And how can companies benefit from it?
In this blog post, we’ll cover the ins and outs of automated English speaking tests. We’ll show you how Versant tests work and why you should consider them to help your organization hire people with high level English language skills.
What is automated English testing?
Automated assessments use (AI) Artificial Intelligence and Machine Learning technologies to evaluate the test taker’s language proficiency. They are highly complex, specialized systems. As a result, they can test all four language skills: speaking, listening, reading and writing.
While it may sound futuristic, automated English testing isn’t new. Pearson English’s suite of Versant automated language tests was launched 25 years ago. Since then, we have delivered and scored over 350 million tests. And Versant isn’t the only test in our portfolio that’s scored automatically. Our Benchmark and Level tests use the same technology.
Since automated English level tests can be delivered online, they are highly flexible. Companies and education institutions use them to help scale their hiring and placement programs. They do so because automated scoring has a number of great benefits. For example, it’s faster, cheaper, more scalable and more accurate than using human scorers.
How does automated English testing work?
Developing technology for automated English testing required a lot of research. We gathered huge amounts of data from human scorers. Then, we trained our Artificial Intelligence and Machine Learning system to score tests the same way that human markers do. So we haven’t replaced human judgment. Instead, we taught a machine to replicate it in a fast and efficient way.
You may be wondering: can machines really learn to evaluate speech like humans do? The answer is yes. But let’s dig a little deeper into how it works.
How are automated English speaking tests developed?
Most of us have had frustrating experiences with speech recognition technology. For example, have you ever called your bank and had to repeat yourself several times (sometimes even shouting), only for a machine to tell you that it can’t understand what you’re saying?
Our technology is different for two reasons. Firstly, we developed it for a specific purpose: the evaluation of English language proficiency. And, we applied very specific, patented optimizations. These make our technology suitable for accurate speech testing.
For example, we developed two kinds of acoustic models, or different models for how speakers pronounce words in certain word sequences:
- Acoustic model for native speakers: this allows our technology to understand how native speakers speak.
- Acoustic model for non-native speakers: this allows the system to be forgiving of mispronunciations, but also recognize them as errors. For example, if a test taker mispronounces a word, our system will recognize the word they attempted to say, and give them an appropriate vocabulary score. However, it will also recognize that they mispronounced the word, which will be reflected in their pronunciation score.
Secondly, our speech recognition technology is embedded in a sophisticated and specialized testing system. This system is designed to pinpoint a test taker’s language proficiency score.
How are automated English speaking tests scored?
To understand what makes our system reliable, it’s important to understand our method of scoring.
Our tests are able to collect an incredible amount of data. When a test taker responds to a question, the system captures their response as a sound file. Our technology then processes this sound file. It analyzes every utterance in an incredible amount of detail.
To give you an idea: we take 10-20 different measurements from a typical response. Over the course of an entire test, we extract over 2,000 pieces of data that will inform the test taker’s final score.
When it comes to scoring English speaking tests, our technology analyzes two different aspects of speech:
The content of speech The matter of speech
The content of speech refers to what the test taker said. Did they respond correctly to the question? Did they say it in a grammatically correct sequence? Did they select the right lexical items and use them appropriately? This information feeds into two subscores: vocabulary and sentence mastery.
On the other hand, the manner of speech refers to how the test taker said what they said. How long did it take the test taker to respond? How often did they pause and where did they pause? Did they pause in the same place that a native or high-proficiency speaker would pause? Did they articulate and enunciate each word with the right intelligibility? This information feeds into three subscores: pronunciation, fluency, and intelligibility.
This method makes our scoring system robust, reliable, and accurate.
What are the benefits of automated English speaking tests?
Now that you understand how automated English speaking tests work, let’s take a look at some of their benefits for organizations.
1. Automated speech testing is accurate and reliable
Thanks to the scoring method we described above, administrators can trust our technology to accurately assess test takers’ English language proficiency. In fact, machines are known to be more accurate scorers than humans: they never get tired and they don’t make mistakes.
2. Automated testing is highly secure
There’s no need to worry about cheating when delivering automated tests. The Versant portfolio has an optional extra layer of security called remote monitoring. This means that our system monitors video, image, and browser movements of test takers. We use an AI detection system to automatically flag suspicious behavior and report it to administrators.
3. AI-scored English speaking tests support diversity
Another huge benefit of using automated tests is that they promote diversity and inclusion. Machines can evaluate test taker performance in a completely impartial, bias-free way. Unlike human scorers, our AI system provides consistent, standardized scores, no matter who’s taking the test.
4. Automated testing is fast and scalable
Speed is another important factor when it comes to assessing language skills. Companies and schools need to make hiring and placement decisions as quickly as possible. With automated tests like Versant, Benchmark, and Level, organizations can get an instant score turnaround. They can test hundreds, potentially thousands of students, and get reliable results within minutes.
5. Automated English level tests are cost-effective
Because automated tests save time, they are more cost-effective than other assessment options. They don’t require extra time from human markers. And, there’s no need to spend money on training or equipment.
Finally, here’s what a valued customer said about Versant:
“Backed with accurate and unmatched technology, the Versant tests helped us standardize & scale up our recruitment process. This led to significant improvement in the quality of hires on communication skills which considerably helped us optimize both our recruitment & training bandwidth.” – Abraham Zachariah, the Global Head of Learning and Development for Cognizant BPS
If your organization needs a fast and scalable solution to English speech testing, learn more about Versant and contact one of our advisors.