Dataset Viewer
Auto-converted to Parquet Duplicate
video
video
6.04
6.04
fidelity
int64
1
3
strategy
stringclasses
3 values
variant
stringclasses
4 values
transcript
stringlengths
21
106
1
modeling
no_joint_attention
Parent: Look, a big red bear.
1
mand_model
no_joint_attention
Parent: Is this a dog or a cat?
1
time_delay
no_joint_attention
Parent: Brown bear, brown bear, what do you see?
1
mand_model
no_joint_attention
Parent: Is this a zebra or a tiger?
1
mand_model
no_joint_attention
Parent: Which one is the puppy?
1
mand_model
no_joint_attention
Parent: What is that?
1
time_delay
no_joint_attention
Parent: Brown bear, brown bear, what do you see?
1
mand_model
no_joint_attention
Parent: Look, a big red pig.
1
time_delay
no_joint_attention
Parent: Do you see the bear?
1
modeling
no_joint_attention
Parent: Brown bear, brown bear, what do you...
2
mand_model
rushed_no_wait
Parent: Look, a big red bear. | Parent: What color is the bear?
2
mand_model
rushed_no_wait
Parent: Is this the little blue truck or the big red truck? | Parent: What does the little blue truck say?
2
modeling
rushed_no_wait
Parent: Look, a brown bear! | Parent: What does the brown bear say?
2
modeling
rushed_no_wait
Parent: Do you see the bear? | Parent: What's the bear wearing?
2
mand_model
rushed_no_wait
Parent: Which one is the mouse? | Parent: Where is the mouse?
2
mand_model
rushed_no_wait
Parent: Where is the bear? | Parent: What does the bear say?
2
time_delay
rushed_no_wait
Parent: Do you see the puppy? | Parent: What's the puppy's name?
2
modeling
rushed_no_wait
Parent: Do you see the mouse? | Parent: Can you find the big bear?
2
mand_model
rushed_no_wait
Parent: Where is the dog? | Parent: Where's the dog?
2
time_delay
rushed_no_wait
Parent: What is that? | Parent: What's that animal?
3
mand_model
responded_not_acknowledged
Parent: Who is that? | Kid: Bear!
3
mand_model
responded_not_acknowledged
Parent: Do you see the mouse? | Kid: Moo | Parent: Where is the mouse?
3
time_delay
responded_not_acknowledged
Parent: Do you see the puppy? | Kid: Puppy! | Parent: What's the puppy doing?
3
time_delay
responded_not_acknowledged
Parent: What's that? | Kid: Bee! | Parent: The bee is flying.
3
mand_model
responded_not_acknowledged
Parent: Which one is the bear? | Kid: Bear | Parent: What does the bear say?
3
modeling
no_response_no_reprompt
Parent: Look, a big red ball. | Parent: Next page.
3
time_delay
no_response_no_reprompt
Parent: What animal is this? | Parent: Okay, next page.
3
modeling
no_response_no_reprompt
Parent: What animal is this? | Parent: Okay, let's look at the next page.
3
mand_model
no_response_no_reprompt
Parent: Is this the truck or the tractor? | Parent: Okay, next page.
3
mand_model
no_response_no_reprompt
Parent: Where is the cat? | Parent: Next page.

ASD Teaching-Fidelity — Synthetic Video Samples (for review)

These are AI-generated (fully synthetic, no real individuals) short video clips depicting a parent running a naturalistic language-teaching episode with a young child around a picture book. They are samples for reviewer feedback on whether the videos correctly portray each teaching-fidelity level.

Each clip is ~6 seconds, 1216×704, with generated audio (dialogue + room tone).

Fidelity levels shown

The parent uses one of three teaching strategies (mand_model = ask then model, time_delay = cue then wait, modeling = just demonstrate). Fidelity = how correctly the parent delivers it:

Fidelity Meaning What you should see
1 — no joint attention Parent prompts, child not attending Child disengaged (looks at own hands/sleeve/hair), no shared focus, no response
2 — rushed (no wait) Child engaged, but parent does not wait ≥3s Parent fires the next line almost immediately, no expectant pause
3 — waited, no acknowledgment Parent waited ≥3s but follow-through missing Two valid forms: child responds but parent does not acknowledge it (responded_not_acknowledged); or child stays silent and parent does not re-prompt (no_response_no_reprompt)

Note on fidelity 3: about half of the fid3 clips have no child response — that is correct. fid3 requires that the parent waited but did not acknowledge or re-prompt; a child response is not required. The distinction from fidelity 1 is that in fid3 the child is attending (joint attention present) whereas in fid1 the child is disengaged.

Contents

  • videos/ — 30 clips: 10 fidelity-1, 10 fidelity-2, 10 fidelity-3 (5 responded_not_acknowledged + 5 no_response_no_reprompt)
  • metadata.csv — per-clip fidelity, strategy, variant, and transcript
  • samples.json — same metadata plus the full generation prompt (scene description) per clip

File names encode the label: fid<L>__<strategy>__<id>.mp4.

How they were made

Text-to-video with LTX-2 (LTX-2.3 22B distilled, HQ two-stage pipeline), a static-camera control LoRA, from gate-verified prompts. No reference image. People and settings are entirely synthetic.

Downloads last month
32