Chris Berube — Craptions

In one sentence

"Craptions" are the inaccurate, poorly formatted, and often inaccessible closed captions that fail to meet the needs of the deaf and hard of hearing community while becoming a standard aesthetic for hearing viewers.

Why read this (and who it's for)

This summary explores the hidden infrastructure of accessibility. It is for content creators, audio engineers, and anyone who has ever struggled to hear dialogue in a loud movie. You will learn why "mumbly" acting is a byproduct of technology, why AI-generated captions currently fail the "human" test, and the legal/social battle to make the digital world more inclusive.

The big idea

The "Craptions" phenomenon highlights a massive gap between the availability of a tool and the quality of its execution. While closed captioning is a vital accessibility requirement under the Americans with Disabilities Act (ADA), the rapid shift to streaming and social media has outpaced the standards required to make those captions usable.

The core of the problem is two-fold: technical and systemic. Technically, modern "naturalistic" acting and high-dynamic-range mixing (where explosions drown out voices) make it harder for machines and humans alike to parse dialogue. Systemically, the sheer volume of content on platforms like YouTube makes manual captioning—the "gold standard"—economically difficult for individual creators, leading to a reliance on "crappy" automated systems that often fail the very people they are meant to serve.

Key ideas

The Definition of "Craptions"

• What it is — "Craptions" are captions that are inaccurate, fail to differentiate between speakers, lack proper punctuation, and are not grammatically correct.
• Why it matters — For the 15% of the US population that is deaf or hard of hearing, "crappy" captions aren't just a nuisance; they are a barrier to entry that forces users to stop watching content entirely.
• In the book — Linda Besner describes them as "halfway there" captions that fail to provide a cohesive experience for the viewer.

The "Social Place" Argument

• What it is — The legal argument that streaming services like Netflix are "social places" and therefore must comply with the ADA.
• Why it matters — It shifts the responsibility of accessibility from physical locations (like theaters) to digital spaces where families and communities interact.
• In the book — The National Association of the Deaf (NAD) successfully argued that a family watching a show together is a social experience that requires equal access for all members.

The Dynamic Range/Naturalism Trap

• What it is — A technical conflict where "naturalistic" acting (mumbling) and high-dynamic-range mixing (loud explosions) make dialogue harder to hear.
• Why it matters — It explains why even hearing people struggle to understand modern films, and why it makes "downmixing" for smaller speakers so muddy.
• In the book — Audio editor Austin Olivia Kendrick explains that as wireless mics gave actors more freedom, they stopped "projecting," leading to performances that are harder to capture and clarify.

The AI Gap

• What it is — The reliance on Google’s universal speech model to generate captions because human captioning is too expensive for most creators.
• Why it matters — While AI is getting better at word recognition, it still struggles with grammar, names, places, and "kinetic" timing (showing words one by one in a way that is hard to follow).
• In the book — Rikki Poynter notes that while AI can handle a dry sentence about dental floss, it fails to capture the nuance of human conversation.

The Cost of Quality

• What it is — The economic barrier where professional captioning can cost $40–$50 per video, which is prohibitive for many independent creators.
• Why it matters — This creates a "voluntary" accessibility model where creators only caption when they have the time or money, leading to inconsistent experiences for users.
• In the book — The "No More Craptions" campaign was a response to this, attempting to turn human-produced captions into a community standard.

Frameworks & tools

The "Gold Standard" of Captioning

To ensure a caption is actually accessible, it must meet these criteria:

1. Speaker Identification: Clearly label who is speaking (e.g., "Linda:").
2. Grammatical Integrity: Use proper capitalization, question marks, and commas.
3. Synchronous Timing: Ensure the text appears and disappears in sync with the audio.
4. Placement: Ensure the text is properly placed on screen so it doesn't obscure the visual action.

The "Downmix" Awareness

When producing audio for different platforms, remember the 128-to-2 Rule:

1. Identify the Source: Recognize that a 128-channel studio mix will be compressed into 2 (stereo) or 1 (mono) channels for most listeners.
2. Prioritize Dialogue: Ensure the "dialogue floor" is high enough to survive the compression of a downmix.
3. Check the "Mumble" Factor: Verify that the actor's performance remains intelligible when the dynamic range is flattened for mobile speakers.

Memorable examples & stories

The "Shitty Gatsworth" Error

During the announcement of Ncuti Gatwa as the new Doctor Who lead, YouTube's auto-captions replaced his name with a swear word. Lesson: AI still lacks the cultural and contextual awareness to handle names and identities correctly.

The "War! Canada" Incident

In Black Panther, the fictional country of Wakanda was captioned as "War! Canada." Lesson: Automated systems often fail spectacularly when encountering unique proper nouns or invented places.

The "Zebra" Confusion

Deaf YouTuber Rikki Poynter watched a beauty video where the auto-captions replaced the word "concealers" with "Zebra," the mammal. Lesson: Contextual errors in AI can make the content completely incomprehensible to the intended audience.

The "Crocamole" Captioning Marathon

Emma Healey describes the grueling process of captioning a children's show, where she had to follow a 75-page technical manual for a minimum-wage job. Lesson: Professional captioning is a high-effort, highly regulated craft that is often undervalued.

Key quotes

"Craptions are captions that are crappy." — Linda Besner Why it matters: It highlights the lack of a professional standard for "good enough" in the digital age.

"There’s no excuse for not providing a transcript... Nearly all podcasts are recorded, right? They’re not live." — Howard Rosenblum (Interpreter) Why it matters: It emphasizes that for recorded media, accessibility is a choice, not a technical limitation.

"I don’t want to wait years and years and years and years for that to happen." — Rikki Poynter Why it matters: It captures the urgency of the disabled community; "eventual" accessibility is not the same as "current" accessibility.

How to apply it

This week

• Audit your watch list: Turn on captions for a show you usually watch without them. Note where the "craptions" fail (missing punctuation, overlapping text, or incorrect words).
• Check your social media: If you post videos on TikTok or Instagram, ensure you are using the "closed" captioning features rather than just "open" aesthetic captions.

Ongoing

• Advocate for "Human-in-the-loop": When using AI tools for captions, always do a manual pass to check for names, places, and emotional tone.
• Prioritize Dialogue Clarity: If you are a creator, remember that "naturalistic" mumbling is harder to caption. Ensure your actors are projecting enough to be captured by the software.

Common pitfalls

• The "AI will fix it" Fallacy: Many believe that because AI is getting better at passing the bar exam, it will soon "solve" captions. However, the "kinetic" way AI displays words (one at a time) can still be harder to read than human-paced blocks of text.
• The "Social Place" Oversight: Assuming that because a platform isn't a "building," it doesn't need to be accessible. The ADA applies to any space where the public gathers.

Connects to

• The "Digital Divide": How accessibility is often treated as a "luxury" feature rather than a fundamental right.
• Human-Computer Interaction (HCI): The study of how humans interact with automated systems, specifically where "predictive text" and "autocorrect" logic can fail to understand human intent.

If you remember only three things

• Craptions are the standard of "good enough" that actually fails the people who need it most.
• Technical limitations (dynamic range, downmixing, and naturalistic acting) make it harder for everyone to hear dialogue in modern media.
• Accessibility is a social right; digital spaces are "places" where people with disabilities deserve equal access to content.

Internalize it

• Reflection questions
  • When I watch a video with auto-captions, do I find them easy to read, or do I find myself constantly re-reading lines?
  • How does the "mumbly" style of acting affect my own ability to consume content on mobile devices?
  • If I were a content creator, what would be the biggest barrier to providing 100% accurate human captions?
• Flashcards Q: What is the "Gold Standard" for closed captioning? A: Captions that identify speakers, use proper punctuation/capitalization, and are perfectly synchronized with the audio. Q: Why is "naturalistic" acting a challenge for captioning? A: It often involves mumbling or overlapping speech, which is harder for AI and humans to parse, especially when downmixed for smaller speakers. Q: What is the "Social Place" argument regarding the ADA? A: The argument that streaming services and social platforms are public accommodations and must provide equal access for deaf and hard of hearing users. Q: Why is AI-generated captioning still considered "crappy"? A: It struggles with grammar, names, places, and the "kinetic" timing of how words appear on screen. Q: What is the "Dynamic Range" issue in movie audio? A: Balancing loud sounds (explosions) with quiet sounds (dialogue) often results in the dialogue being drowned out or lost during the downmixing process.

Fidelity note

This summary is based strictly on the provided transcript. I am confident in the technical explanations regarding "downmixing" and "dynamic range" as they were directly attributed to the audio experts in the text. The distinction between "open" and "closed" captions is also clearly defined in the transcript. The summary does not include any information regarding the specific "30% increase" in manual captions beyond what was stated as a YouTube statistic.