Readability is not a new term. But, it has become something  of a buzzword recently in the lives of digital professionals, especially those who work with the written word.

Ernest Hemingway was a huge proponent of readability in fiction, but reliability obsesses Silicon Valley too.

Readability is essentially how easy a piece of text is to understand. Different parts of writing contribute to its overall readability including:

• Word choice

• Sentence length

• Use of passive voice

• Adverbs 

What is a readability score?

A readability score is a grade or number that tells you how easy (or hard) your writing will be for someone to understand. 

A high readability doesn’t mean a piece of writing cannot be understood, or that it is hard to understand necessarily. However, a piece of writing with a lower readability will be understandable by more people.

If your writing has a high readability score it is likely that people will have to take more time unpicking longer sentences and clauses within clauses, rather than focusing on the content of what you’re writing. 

Why readability matters for AI tutors

There are several reasons why readability is important to the training of large language models (LLM) - huge language databases used to train AIs in natural-sounding, conversational human language.

Some of the most common are:

• Data quality — as an AI Tutor you will see an emphasis on data quality related to readability, this is because clearer language helps train AI models more effectively. Badly written text samples can lead to AI misunderstandings and unhelpful recommendations

• Bias — well-structured data helps the model avoid bias in training data. This helps us create fair and unbiased models

• Efficiency — clearly written texts save time and computer power when processing data

• Evaluation — AI developers and researchers can work more effectively when data is readable. It helps them understand the model’s responses and debug it when errors occur

Editors’ point of view

Unfortunately, as an editor, I noticed writers can often miss the readability requirement. Many tasks I edited were written at a postgraduate level. While readability was an important consideration for the project, I felt most of my time was spent editing long unwieldy sentences. While the other project requirements were met most of the time, poor readability has meant more time spent on the same task. In some projects, there is no possibility for writers to re-do tasks, so it’s always important to hit requirements the first time.

By creating highly readable text we contribute to more effective training of AI models.

Why readability is specifically important for writers

You might have been wondering “Why is this important to AI tutors, sure readability is important in general, but how does it apply?”

Here’s the crux. AI chatbots can be and are used by a variety of different ages and education levels. In the same way that webmasters, copywriters and content managers try to create scannable and easy-to-read content, those working in AI, especially those training AI language models, must do our best to make AI responses as natural and human-sounding as possible.

Some projects have readability requirements. One such project I worked on was the Complex Reasoning project. 

Here, one of our aims was to write at a Grade 10 comprehension - equivalent to high school level, 15 or 16 years of age.

Here’s an example of post-doctorate level writing vs a sample of writing at a Grade 10 reading age from Hemingway.

Post doctorate reading comprehension

Grade 10 reading comprehension

How editors and writers can work together

Writing and editing are different sides of the same coin. Both are distinct skills. However, both groups are working towards a shared goal. So, it is important to collaborate effectively and support each other.

This can be achieved by understanding each other’s role. Spend time reading each other’s project guidelines, see what each other are talking about on Discord channels and always be sure to leave positive and constructive feedback for good work or when misunderstandings occur.

Readability tools

Readability scores are generated by readability tools. There are a whole host of AI-generated readability tools that analyse a piece of writing and give it a score based on the reading comprehension needed to understand it.

Let’s explore some of the most common readability tools. 

Hemingway app

Known for his direct style characterised by short sentences and active language, Hemingway became the inspiration for the Hemingway App. He said:

“If I started to write elaborately, or like someone introducing or presenting something, I found that I could cut that scrollwork or ornament out and throw it away and start with the first true simple declarative sentence I had written.”

The app is primarily a readability checker and is great for short and punchy pieces of text.

The app grades a piece of writing based on readability levels. The lower the grade the more understandable a writing sample is. Grade 9 is the adult average in the United States, so the app encourages you to refine your writing until it reaches this level.

Making use of a highlighter, the app brings very hard-to-read sentences to your attention with a red pen –these are the areas you should address first. Hard-to-read sentences are in yellow, ‘weakeners’ in blue and word choices with a simpler alternative in purple.

Grammarly

Grammarly is another web app that assesses readability and gives suggestions to improve writing.

While it is primarily a grammar checker, it does give insights into readability. Grammarly can help you refine longer pieces of content with detailed grammar checks. Users can upload entire documents easily and get in-depth feedback on grammar and readability.

Grammarly can be added to Google Chrome as an extension, used in word, or online via its browser app. 

Rather than giving a reading comprehension grade as an age, Grammarly gives writing a score out of 100. This is its performance score. The higher the score, the better. Scores can be increased by addressing the suggestions for improvement that the AI tool makes.

Above, is an insight into the Grammarly app using a sample of writing from this very blog. Have a look at this screenshot. You can see my writing received a score of 83/100. It helpfully suggests that I could improve my writing by adding hyphens between ‘hard to read’ and that I should add an article so that ‘red pen’ becomes ‘a red pen’. 

Elements of its readability metrics include word length and sentence length. Grammarly uses the Flesch Reading Ease scoring system where higher scores are better - it also gives you a summary of how easy your text is to read. For example:

“Your text is likely to be understood by a reader who has at least some college education, but it may not be easy to read.”

Training AI for the future 

The role of AI is going to evolve, ethical training and bias mitigation are becoming increasingly important to ensure AI tools follow human values and legal standards. Readability is at the core of this paradigm shift.

An AI tutor may find they’re increasingly tackling specialised tasks that address ethical and legal challenges within specific domains, making the work both impactful and intellectually stimulating.

Being part of this process means contributing to the next big advancement in AI. Qualifying as an AI tutor is an opportunity to make a significant impact, helping to develop AI that is both ethical and fair. This work is not only rewarding but also essential to the positive evolution of AI.