I'm going to lay out the actual numbers for a real project structure. Not estimates. Not industry averages. A specific 100-episode short drama project, translated from Chinese to English, with the three standard approaches priced out line by line.
The reason I'm doing this is because most of the 'MTPE vs. human translation' cost comparisons I see in the industry are either so abstract they're meaningless or so optimistic about MTPE that they might as well be sales pitches. I want to show the math with enough specificity that you can plug in your own rates and see what happens.
The project parameters: 100 episodes, averaging 2.5 minutes each. Dialogue density of roughly 180 words per episode in Chinese source. That's 18,000 source words across the full project. Target language: English. Content type: contemporary short drama — workplace/romance genre, moderate cultural density, some idiomatic dialogue, no period-specific register.
Let me walk through each approach.
Approach 1: Pure human translation
Rate: $0.08-0.12 per source word for Chinese-to-English subtitle translation with timing and QA. Let's use $0.10 as the midpoint.
18,000 words at $0.10 = $1,800 for the initial translation.
But that's not the full cost. Subtitle translation includes timing, which adds roughly 30% to the per-word cost. So the actual translation with timing: $2,340.
Linguistic QA pass: typically 25-30% of the translation cost. Let's call it 28%. $655.
Revision and corrections from QA: roughly 10% of translation cost. $234.
Project management and coordination for a 100-episode project across multiple translators working in parallel: roughly 15% of total. $484.
Total pure human: $3,713 per 100-episode batch. At that rate, a 100-episode project with 3 target languages = $11,139.
Timeline: A competent subtitle translator handles about 20-25 minutes of runtime per day. 100 episodes at 2.5 minutes each = 250 minutes. With 3 translators working in parallel: roughly 3.5-4 working days for translation, plus 1 day for QA and corrections. About one week for a single language pair. Three language pairs running in parallel: still about one week if you have the team, three weeks if you don't.
Quality: Assuming competent translators with short drama experience, expect a quality score of 88-94 out of 100 on a composite metric covering grammatical accuracy, character voice consistency, cultural adaptation, and subtitle timing compliance. I'm using the same scoring framework I described in the MTPE strategy piece — 20% grammar, 25% character voice, 25% cultural adaptation, 15% timing, 15% overall readability.
Approach 2: Pure machine translation
MT processing cost: depends on the engine. A cloud-based neural MT API (Google, DeepL, Microsoft) charges roughly $20-40 per million characters for the standard tier. 18,000 Chinese source words at roughly 2 characters per word (including punctuation): roughly 36,000 characters. Processing cost: under $2. Let's round up and call it $5 because you'll run it more than once for testing.
Subtitling and timing: you still need someone to create the subtitle files from the MT output and time them to the video. This is less skilled work than translation — it's essentially data entry with timing software. Rate: roughly $0.02-0.03 per source word. 18,000 words at $0.025 = $450.
Total pure MT: $455 per 100-episode batch. Three languages: $1,365.
Timeline: MT processing takes minutes. The subtitling/timing work is the bottleneck. A subtitler doing pure timing (no translation) can handle roughly 60-80 minutes per day. 250 minutes: about 3-4 working days for timing. Total project: 4-5 days for a single language pair.
Quality: On short drama content, expect 45-58 out of 100 using the composite metric. Grammar will be mostly fine (modern neural MT is grammatically competent). Character voice will be terrible (20-30/100). Cultural adaptation will be terrible (15-25/100). The output will be informationally accurate in the sense that a viewer can follow the plot, but the emotional and character-specific dimensions of the dialogue will be flattened.
This is the thing about MT quality on creative content that the cost-per-word comparisons always miss. The MT output isn't 'mostly good with a few errors.' It's 'structurally deficient in the dimensions that matter most for the content type.' The 15-25/100 cultural adaptation score isn't a small deficiency. It's a category failure. You're not saving money by using MT. You're spending $455 to produce content that will damage the show's reception.
Approach 3: MTPE with tiered editing
MT processing: $5 (same as above).
Post-editing rate: this is where it gets interesting. MTPE rates are typically 40-60% of the pure human translation rate, depending on the MT quality and the content type. For Chinese-to-English short drama, with the MT quality I described above, the effective post-editing rate is roughly $0.05-0.06 per source word. Let's use $0.055.
But that rate applies to the full word count, and the whole point of the tiered system is that you're not editing the full word count with equal intensity. In practice, a tiered MTPE workflow has three effective rates:
Tier 1 (no-touch, ~45% of lines): essentially zero editing cost. The post-editor skims for confirmation. Effective per-word cost: near $0. But you're still paying the post-editor for their time, so let's assign a nominal $0.01/word for the skim.
Tier 2 (light edit, ~35% of lines): the post-editor is making register adjustments, compressing for timing, sharpening emotional tone. Effective per-word cost: roughly $0.05/word.
Tier 3 (deep edit, ~20% of lines): essentially retranslation. Effective per-word cost: roughly $0.09/word — close to the pure human rate, because the post-editor is doing comparable work.
Blended rate across all three tiers: roughly $0.04/word. 18,000 words at $0.04 = $720.
Wait. That's lower than the $0.055 I quoted. Here's why: the $0.055 is the rate you'd pay if the post-editor were editing every line with moderate intensity. The tiered system reduces the effective rate because the post-editor is spending proportionally less time on the content that doesn't need editing. If you're paying by the hour instead of by the word, the math works out differently, but the proportional savings are similar.
Timing: included in the post-editing workflow because the post-editor works within the subtitle template. $0.
QA pass: still needed, but lighter than the pure human QA because the post-editor has already done one round of quality filtering. Roughly 18% of the post-editing cost. $130.
Project management: slightly lower than pure human because the MTPE workflow is more standardized. 12% of total. $103.
Total MTPE: $958 per 100-episode batch. Three languages: $2,874.
Timeline: The post-editor's throughput with the tiered system is roughly 40-50 minutes of runtime per day — faster than pure human because they're not translating from scratch, but slower than pure MT timing because they're actually editing. 250 minutes: about 5-6 working days. With QA and corrections: roughly 7 working days for a single language pair.
Quality: 82-89 out of 100 on the composite metric. The gap from pure human quality is concentrated in character voice and cultural adaptation. Tier 3 deep-edit lines will be close to human quality. Tier 2 light-edit lines will be solid but occasionally lack the specificity that a full human translation would provide. Tier 1 no-touch lines will be fine because they're informational content where MT is adequate.
The comparison
Here's the summary for a single language pair, 100 episodes:
Pure human: $3,713 | 5-7 working days | 88-94 quality
Pure MT: $455 | 4-5 working days | 45-58 quality
MTPE (tiered): $958 | 7 working days | 82-89 quality
The cost-quality tradeoff isn't linear. Going from pure MT ($455) to MTPE ($958) costs an additional $503 but gains 24-31 quality points. Going from MTPE ($958) to pure human ($3,713) costs an additional $2,755 but gains only 5-12 quality points. The MTPE approach captures roughly 85-95% of the pure human quality at roughly 26% of the pure human cost premium.
Another way to look at it: MTPE costs 74% less than pure human while delivering quality that's within 6-12 points. Pure MT costs 88% less but delivers quality that's 30-43 points lower. The marginal cost of quality improvement from pure MT to MTPE is enormously more efficient than the marginal cost from MTPE to pure human.
This is the financial case for MTPE. Not 'MTPE is cheaper.' It's: the quality improvement per dollar spent is highest in the MTPE tier, and the marginal return on spending beyond MTPE diminishes sharply.
The numbers change with content type and language pair
Everything above is for Chinese-to-English contemporary short drama. The math shifts significantly with different parameters, and I want to be honest about where MTPE's advantage narrows or disappears.
Historical or period drama: The MT failure rate on period-specific register and terminology pushes the tier 3 proportion from 20% to 40-50%. The blended rate shifts from $0.04/word to $0.06-0.07/word. MTPE total for 100 episodes: $1,280-1,450. Still cheaper than pure human. But the quality gap between MTPE and pure human narrows to 3-8 points because pure human translators handle period register much more reliably than post-editors working from MT output that's already wrong on register. The financial case is weaker.
Low-resource language pairs: If the MT engine's training data is thin for the language pair, the MT output quality drops and the post-editor is fixing grammar errors rather than editing for dramatic effectiveness. The tier 1 proportion drops from 45% to maybe 15-20%. The tier 3 proportion rises to 40%+. The post-editor is essentially translating from scratch on most lines while also having to evaluate and discard the MT output first, which makes MTPE slower than pure human translation. Chinese-to-Thai, for example, is a language pair where I've seen MTPE take 15-20% longer than pure human because the MT quality is low enough that the evaluation step is waste rather than efficiency.
High-volume, low-complexity content: The flip side. If the short drama content is primarily informational — a workplace procedural with minimal emotional dialogue, no idiomatic complexity, and straightforward sentence structure — the MT quality is higher and the tier 1 proportion can reach 60-65%. MTPE cost drops to roughly $650 per 100 episodes. Quality is within 3-5 points of pure human. This is the sweet spot for MTPE, and it's where the cost advantage is largest.
The point is: MTPE isn't a universal cost solution. It's a cost solution that works best for content types and language pairs where MT quality is high enough to provide a usable editing base and cultural complexity is moderate enough that the tier 2/3 proportions stay manageable. Run the math on your specific content before committing to an approach.
The hidden cost that most comparisons ignore
There's a cost that doesn't show up in any of the per-word calculations above: the cost of bad subtitles to the platform.
A short drama with flat, characterless subtitles doesn't generate the word-of-mouth and social media engagement that drives viewer acquisition. A viewer who watches a subtitled episode and feels that something is off — the dialogue seems bland, the characters all sound the same — doesn't complain about the subtitles. They just don't recommend the show. The cost is invisible because it manifests as absence rather than presence. Lower engagement. Lower retention. Lower organic growth.
I can't put a precise dollar figure on this because the relationship between subtitle quality and viewer behavior is mediated by too many variables. But I've seen enough projects to know that the platforms with consistently good subtitles have measurably higher completion rates and recommendation rates for their international content than the platforms with inconsistently good subtitles. The subtitle quality is not the only variable — content quality matters more — but it's a variable, and it's one that platforms can control.
The pure MT approach saves $500 per 100 episodes per language compared to MTPE. That's real money at scale. But if the quality difference between pure MT (45-58/100) and MTPE (82-89/100) translates into even a 5% difference in viewer retention across the episode catalog, the pure MT savings are offset many times over by the retention cost. I can't prove this with a controlled experiment because nobody runs controlled experiments on subtitle quality and viewer retention. But the directional argument is strong, and the platforms that have invested in subtitle quality seem to agree, because they keep paying for it.
Artlangs Translation provides video MTPE services for short drama content across 230+ language pairs. Tiered editorial workflows (no-touch, light edit, deep edit) calibrated to content type and language pair. Per-project cost modeling so you can compare approaches before committing. Spot-check QA, calibration sessions, and MT engine-specific error logging built into the workflow. If you're trying to find the cost-quality balance point for a high-volume short drama project, the math is where you start — and the math says MTPE for most content types, pure human for the rest, and pure MT almost never.
