Comparing British and American Audio Description of Movies

This paper compares British and American audio description (AD) of movies, which is a narrative technique designed to enable blind and visually impaired individuals to follow the story. The authors note that AD is "far more constrained than most narratives: the descriptions not only need to convey the story in the movie, but they must also fit into gaps between dialogue and they need to conform to guidelines that exist in each region."

The study uses a pre-existing corpus that provides both human-authored American and British audio description for each of 206 movies. The authors motivate specific hypotheses about differences between the two regions based on guidelines written for these two regions, alongside the impressions from a practitioner in the field.

The hypotheses tested are:

1. American AD uses a higher frequency of American English lexical items.

2. British AD uses a higher frequency of British English lexical items.

3. British AD uses the present progressive aspect more frequently.

4. British AD exhibits greater subjectivity in interpreting visual information and character emotions.

5. British AD uses the passive voice more frequently.

6. British AD names characters in advance more frequently.

7. American AD voices over dialogue (and music) more frequently.

The methods involve automatic NLP processing using the Stanza toolkit, and statistical significance testing with paired, two-tailed permutation tests (N=10,000, α=0.05).

The results, shown in Table 2, support all hypotheses:

• American terms: 16.8 per 1,000 sentences in US AD vs. 5.0 in UK AD (p < 0.01)

• British terms: 15.8 per 1,000 sentences in UK AD vs. 4.6 in US AD (p < 0.01)

• Present progressive: 3.0% of sentences in UK AD vs. 0.3% in US AD (p < 0.01)

• Subjectivity: mean sentence score 16.2% in UK AD vs. 13.3% in US AD (p < 0.01)

• Passive: 4.6% of sentences in UK AD vs. 1.3% in US AD (p < 0.01)

• Naming in advance: 31.4% of characters in UK AD vs. 9.0% in US AD (p < 0.01)

• Dialogue overlap: 8.5% of AD time in UK AD vs. 14.1% in US AD (p < 0.01)

Additional findings include:

• The most frequent American terms in American AD are 'SUV', 'flashlight', and 'sidewalk'; the most frequent British terms in British AD are 'amongst', 'car park', and 'windscreen'.

• When a character is named in advance in British AD, they are named on average 11 minutes early, compared to seven minutes in American AD.

• In most cases (59.6%), both versions name the character immediately after they are named in the dialogue.

• For scene changes, a BERT model trained to identify scene boundaries from AD text achieved significantly lower recall on British transcripts (55.4% vs. 72.4% when trained on UK data; 42.9% vs. 66.0% when trained on US data; 55.1% vs. 69.0% when trained on both), suggesting scene boundaries are verbalised less explicitly in British AD than in American AD.

The authors conclude: "We have shown at scale the existence of small but statistically distinguishable differences between how the narrative is verbalised in American vs. British descriptions. All of the differences are justified based on differences between the regions' guidelines, except for lexical variation, which has not received prior theoretical treatment."

Improvements to AI Systems:

1. Region-Aware Audio Description Generation: An AI system for generating audio descriptions can be fine-tuned to produce either American or British AD by incorporating the identified linguistic markers. Specifically, it can:

• Automatically select region-specific lexical items (e.g., SUV vs. 4x4, sidewalk vs. pavement) based on the target audience.

• Adjust verb aspect usage: use present progressive more frequently in British AD (e.g., she is walking vs. she walks).

• Modulate subjectivity: increase interpretive language (e.g., he looks angry vs. he frowns) for British AD, and keep more objective, neutral descriptions for American AD.

• Control passive voice frequency (higher in British AD) and dialogue overlap (lower in British AD, meaning more pauses for description).

• Implement character naming strategies: name characters earlier and more frequently in British AD (on average 11 minutes before dialogue reveals the name) vs. American AD (7 minutes).

2. Style-Transfer for AD Localization: An AI system can take an existing AD script from one region and automatically convert it to the other region’s style. For example:

• Replace American lexical items with British equivalents and vice versa.

• Rewrite sentences to shift from active to passive voice (or reverse) based on target region.

• Add or remove subjective interpretations (e.g., she seems nervous vs. she taps her fingers) to match the target region’s subjectivity level.

• Adjust the timing of character name introductions to align with regional norms.

3. Scene Boundary Detection Improvement: The finding that British AD verbalizes scene changes less explicitly can be used to train a more robust scene-boundary detection model. An AI system can:

• Learn to infer scene boundaries from implicit cues (e.g., changes in character actions, locations, or temporal markers) rather than relying solely on explicit phrases like cut to or now we see.

• Use a dual-region training approach: train on both American and British AD data to improve generalization, but also allow region-specific fine-tuning to match the explicitness level of the target AD style.

4. Automated AD Quality Assessment: An AI evaluator can be built to check whether a generated AD script adheres to regional guidelines by measuring:

• Frequency of region-specific lexical items (using a predefined list).

• Proportion of present progressive, passive voice, and subjective sentences.

• Timing of character name introductions relative to dialogue.

• Degree of dialogue overlap (should be lower for British AD).

This evaluator can provide feedback to AD generators or human editors to ensure consistency with regional norms.

5. Personalized AD for Multilingual/Regional Audiences: An AI system can dynamically adapt AD in real-time for streaming platforms, allowing users to select British style or American style descriptions. The system would:

• Use the identified statistical thresholds (e.g., 3.0% vs. 0.3% present progressive) as target parameters.

• Apply a rule-based or neural post-processing step to adjust the generated text to meet those targets, ensuring the narrative feels natural to the chosen region.

6. Improved NLP Pipelines for AD Corpora: The study’s use of Stanza for POS tagging and dependency parsing can be enhanced by training domain-specific models on AD text. An AI system can:

• Fine-tune a language model on both American and British AD corpora to better handle AD-specific syntax (e.g., frequent use of present progressive, passive constructions) and improve downstream tasks like subjectivity classification or scene boundary detection.

7. Cross-Regional Narrative Understanding: An AI system trained on this contrastive data can learn to identify regional narrative styles in other media (e.g., novels, subtitles). It can:

• Classify the origin of a narrative text based on the linguistic features identified (lexical, aspect, voice, subjectivity, naming, dialogue overlap).

• Use this classification to automatically adapt content for different regional audiences in translation or dubbing tasks.

Narrating the visual component of movies is known as audio description. It is a narrative technique designed to enable blind and visually impaired individuals to follow the story. However, it is far more constrained than most narratives: the descriptions not only need to convey the story in the movie, but they must also fit into gaps between dialogue and they need to conform to guidelines that exist in each region. In this work, we compare audio description created in the United Kingdom against audio description created in the United States. We use guidelines written for these two regions, alongside the impressions from a practitioner in the field, to motivate specific hypotheses about the differences. We test these hypotheses against our pre-existing corpus, which provides both human-authored American and British audio description for each of 206 movies. Results provide quantitative evidence to uphold all tested hypotheses, including differences in lexicon, the use of the progressive aspect, the use of passive constructions, the use of subjective adjectives and modifiers, when characters are named, how scenes are cued, and degree of overlap with movie dialogue and music. Our work offers a quantitative lens into the narrative technique of audio description.