Why You Can’t Predict Exact Words in Real Speech

Many stenography methods quietly assume this:

If you’re skilled enough, you should be able to anticipate what people are going to say.

That assumption feels reasonable.
It’s also wrong because of how human language actually works.

And we know this because of research.


What “Psycholinguistics” Means (In Plain Language)

Psycholinguistics is the study of how the human brain processes spoken language in real time — how we listen, understand, anticipate, adjust, and recover while people are actually talking.

It doesn’t study how language should look on paper.
It studies what the brain can realistically do as speech unfolds, with all its pauses, corrections, and changes of direction.

That distinction matters a lot for reporters.


What Research Shows About Predicting Speech

Researchers often measure predictability using something called a cloze task.

In a cloze task, people are given the beginning of a sentence and asked to guess the exact next word or phrase.

Example:

“I went to the store to buy some ___.”

Most people will guess something reasonable. I would say bananas.
That’s because this sentence is highly predictable. So, bananas, it is.

But real conversation doesn’t work like that.


Scripted Speech vs. Real Speech

Prepared or scripted speech

(read text, rehearsed material, formal writing)

  • Exact word prediction accuracy: about 60–90%

  • Language is fluent, structured, and constrained

  • Few interruptions or course corrections

Spontaneous, unscripted speech

(testimony, depositions, interviews, everyday conversation)

  • Exact word prediction accuracy: about 20–50%

  • Often closer to 30–40%

  • Speech includes:

    • false starts

    • repairs

    • hedges

    • mid-sentence changes

    • unfinished thoughts

In other words:

Humans are not very good at predicting the exact wording of natural speech.

This finding is consistent across decades of research.
It’s not controversial.


What Humans Are Good At Predicting

This part is important.

People are actually quite good at predicting:

  • When a speaker is finishing a thought

  • The general meaning of what’s coming

  • Turn-taking timing

Prediction accuracy for timing and general meaning can be 70–90%.

But that’s not the same thing as predicting verbatim wording.

Conversation is not a fill-in-the-blank exercise.

Even speakers don’t know exactly how they’ll finish a sentence until they do.

And how many times have you hit that question or answer symbol and the attorney or witness decided to keep going?

Often.


Why This Matters for Stenographers

Steno happens right at the point where language is least predictable.

So if a method assumes that you should routinely commit to long, exact phrasing before the words are confirmed, it’s asking your brain to do something humans are not built to do well.

That has consequences.


1. Anticipation creates extra correction work

When you write ahead of confirmed speech, you’re making a prediction.

When that prediction doesn’t land — which is common — you pay for it with:

  • conflict resolution

  • backtracking

  • hesitation

  • cleanup

That cost adds up over time.


2. Phrase-heavy writing quietly increases mental load

Long phrase outlines work best when speech is predictable:

  • boilerplate language

  • repeated formal phrases

  • names and fixed terminology

These resemble scripted speech.

Depositions and testimony don’t.

Treating spontaneous speech like it’s scripted forces your brain to constantly fix prediction errors — and that raises mental load minute by minute.


3. “You should have known what they were going to say” is false

This is the part reporters often internalize.

If exact wording in spontaneous speech is only predictable about one-third of the time, then:

Failure to anticipate exact phrasing is the expected outcome of how language works.


The Design Lesson

This research doesn’t say:

  • “Never brief”

  • “Never phrase”

  • “Never anticipate”

It says something simpler:

Methods that assume high verbatim predictability in natural speech are adding mental load by design.

And unnecessary load always shows up later as:

  • fatigue

  • hesitation

  • late-day drops

  • plateaus

Because the method is fighting the input.


One Sentence to Remember

If you remember nothing else, remember this:

Humans predict meaning and timing well in conversation — but exact wording poorly. Any steno method that assumes otherwise is adding load by design.


Final Thought

If complexity were the answer, the field would be thriving.

It isn’t.

Sometimes the most professional move is to stop asking your brain to do what it was never built to do.

Back to blog