User guide

Getting started with DIALS

For teachers and researchers. Learn the four features DIALS looks for, try a real result, and see how to read what DIALS reports.

What DIALS does

DIALS reads a piece of student writing and finds places where four features of African American English (AAE) appear. It reports how many it found and a density, which is the number of features per word. Density lets you compare samples of different lengths.

Contrastive, not corrective

DIALS labels describe how a sentence differs from Mainstream American English (MAE). They are not error marks and say nothing about the quality of the writing. The MAE versions in this guide are there to show the contrast, not to correct anyone.

The four features

DIALS v1 detects exactly these four features. In the app, each one has a short colored label. Hover over (or tap) any label to see its definition.

SVA

Variable subject-verb agreement

The verb isn't marked to agree with its subject the way MAE marks it.

he / she / it + bare verb, don't, or have → MAE: verb-s, doesn't, has

Counted

He walk to school. → He walks to school.
She don't like math. → She doesn't like math.
Maria have a notebook. → Maria has a notebook.
My pencil need to be sharpened. → My pencil needs to be sharpened.

Not counted

They walk home.: Plural subject, so the bare verb already agrees.
He can go.: A bare verb after a modal (can, will) is expected.
She doesn't like it.: The verb is already marked (doesn't).
He ready.: A missing is, so this is zero copula, not SVA.

COP

Zero copula

A present-tense form of “to be” (am, is, are) is left out.

subject + predicate → MAE: subject + am / is / are + predicate

Counted

He ready. → He is ready.
They outside. → They are outside.
The classroom warm. → The classroom is warm.
He going home. → He is going home.

Not counted

He ready yesterday.: Past tense (was) is outside the v1 feature.
He be ready after practice.: Habitual be is present, not missing.
The ready student: An adjective describing a noun, not a predicate.
Maria backpack: A missing 's, so this is zero possessive.

POS

Zero possessive

Ownership or relationship is clear without a possessive 's.

possessor + possessed noun → MAE: possessor's + possessed noun

Counted

My brother jacket → My brother's jacket
The teacher desk → The teacher's desk
Maria backpack → Maria's backpack
My partner favorite food → My partner's favorite food

Not counted

school bus: A standard noun compound, not ownership.
my friend Kevin: Two names for the same person, not possession.
my favorite food: The possessive (my) is already there.
South Euclid: One multi-word place name.

ART

Indefinite article

“a” is used before a word that starts with a vowel sound.

a + vowel-sound word → MAE: an + vowel-sound word

Counted

a apple → an apple
a elephant → an elephant
a honest mistake → an honest mistake
a 8-year-old → an 8-year-old

Not counted

a university: Starts with a “you” sound, a consonant.
a one-way street: Starts with a “w” sound, a consonant.
an apple: Already uses an.
a banana: Consonant sound, so a is expected.

Try it: a real result

Below is a real DIALS result, recorded from the Analyze page. Click a highlight or a row to see the details DIALS gives for each feature it finds. When you paste your own writing on the Analyze page, you get the same layout. Your own analysis takes about a minute to run.

Words

Accepted

Review

Density

0.1724

SVA1COP2POS1ART1

Original text (click a highlight)

is on the chair. for school but like math. We saw at the zoo. waiting for the bus.

Accepted5

Select a highlight or a row to see the details DIALS records for it. This is the same detail panel you get on the Analyze page.

Reading your results

Words: The number of words in the sample. This is the bottom of the density fraction.
Accepted: Features DIALS is confident about. Only accepted features count toward density.
Pending review: Possible features with mixed evidence. They don't count until a reviewer accepts them.
Density: Accepted features ÷ words. In the example above, 5 features ÷ 29 words = 0.1724, or roughly 17 features per 100 words.
Span: The exact words DIALS flagged.
Trigger: The part of the span the pattern rule matched on.
Rule confidence: From 0 to 1. How strongly the hand-written pattern rule matched.
Model confidence: From 0 to 1. How likely a model trained on annotated writing thinks the match is real.

How DIALS decides

1
Find candidates
Pattern rules scan the text for shapes that look like each feature (for example, “a” followed by a vowel sound). An AI reading of each sentence can confirm, adjust, or add candidates.
2
Score them
Each candidate gets a rule confidence and a model confidence.
3
Make a first call
If both scores are high and nothing is flagged as tricky, the candidate is accepted. If both are low, it is rejected. Anything in between goes to review.
4
Check against the guidelines
For review candidates, two separate AI checks compare the candidate with written linguistic guidelines. If both agree it is the feature, it is accepted. If both agree it isn't, it is rejected. Otherwise it waits for a human reviewer.

Thresholds (for researchers)

Automatic accept: rule confidence ≥ 0.75 and model confidence ≥ 0.75, with no caution flags.
Automatic reject: rule confidence < 0.50 and model confidence < 0.50.
Everything else starts as review and goes through the guideline check.
Some patterns are always routed to review because they are known to be ambiguous, such as noun-noun pairs that could be compounds or possessives.

Analyzing many samples

Prepare a CSV or Excel (.xlsx) file with an id column and a text column. Any other columns are ignored.
Go to Batch Upload, choose the file, and click Upload & analyze.
Each row takes about a minute. You can leave the page and come back later. Your past batches are listed on the right.
When it finishes, click Download CSV to get the full results.

Download a template CSV

What's in the results CSV

One row for each detected feature. The sample's id and text repeat on each of its rows. Samples with no detections get a single row with the feature columns left blank.

feature_type, span, trigger: Which feature was found and the exact words, as on the Analyze page.
outcome: accept, review, or reject.
counted_in_density: Whether this row counts toward the sample's density (accepted items only).
sva_count, cop_count, poss_count, art_count: Accepted totals for each feature in that sample.
total_density: Accepted features ÷ words for that sample.
status: “ok”, or an error message if that row couldn't be analyzed.

Reviewing candidates

Reviewer accounts see a Review Queue listing every candidate still pending review. Each card shows the flagged span, the sentence it came from, and both confidence scores. Add an optional note, then choose:

Accept: It is the feature. It now counts toward that sample's density.
Reject: It is not the feature. It never counts.
Needs discussion: You're unsure. It leaves the queue for the team to discuss and does not count.

To get reviewer access, check “I am a reviewer” when you sign up, or ask the project team to update your account.

Limits & FAQ

Does DIALS detect other AAE features, like habitual be?

Not yet. Version 1 covers only the four features above. Anything else is ignored and does not affect density.

Why wasn't a missing “was” or “were” counted?

Zero copula in v1 covers present tense only (am, is, are). Past-tense omissions are out of scope.

How long does an analysis take?

About a minute per sample, because each one goes through several checks. Batches run in the background.

Is there a size limit?

Batch files can have up to 100 rows.

Can DIALS be wrong?

Yes. That's why uncertain cases go to review instead of being counted. Treat density as a research measure, not a grade.