Use this article to understand how Smartling counts words, how fuzzy matches reduce cost, how cost estimates are calculated, and why an estimate can differ from your final word count.
​

In this article

Word counts represent the sum of words in source strings, excluding tags and placeholders. They represent the quantity of work done by translators, editors and reviewers in a Smartling account over a particular time period.

Word count reports are intended for tracking and invoicing of work completed. The counts reflect the number of source words and weighted words handled by each user and are broken out by language, workflow step type, and fuzzy tier. SmartMatch information does not appear on word count reports, since it does not constitute work done by users.

The following sample Word Count report shows two rows of work on the same job into French (France): a translation row with a Fuzzy Match Profile and a fuzzy tier, and an edit row with neither.

The Word Count report is available in the Smartling dashboard, where it can also be downloaded in CSV format. Agencies can download reports for anything their team worked on, and individual linguists can download word count reports on their own work.
​

Like word counts, character counts represent the total number of characters in source strings, excluding tags and placeholders. Character counts are particularly useful for measuring the volume of translation, editing, and review work for languages such as Chinese and Japanese, which do not use spaces to separate words.

How your files are broken down into strings can affect the character count. For example, certain file directives can cause content to be ingested as a placeholder. The following table explains how each type of content is counted:

When calculating the size of a translation job, it’s customary in the translation industry to account for previous translations available to the translator when the work is performed.

For example, translating the string “Once upon a time” requires less work if the translation memory (TM) already contains a translation for the similar string “Once upon a time,” (with a comma) than if the string had to be translated from scratch. A pre-existing TM entry that is similar to the source string being translated is called a fuzzy match.

The similarity between two strings is measured using a method called Levenshtein distance, an industry-standard way of calculating how much two strings differ. In Smartling, the similarity of a fuzzy match is expressed as a percentage called a fuzzy score. Higher percentages indicate greater similarity.

For example, the strings “Once upon a time” and “Once upon a time,” have a fuzzy score above 90%. A high fuzzy score indicates that the TM already contains a translation for a string very similar to the one being translated, which usually reduces translation costs. The amount of the discount depends on the Fuzzy Match Profile that applies to the workflow step.

Translation pricing is typically based on the number of source words translated and the fuzzy scores of the strings. A Fuzzy Match Profile groups fuzzy scores into pricing bands, or tiers, with higher-score tiers generally receiving greater discounts.

The following table shows an example of fuzzy tiers. The tiers and discounts configured for your account may differ.

A set of fuzzy pricing tiers, like the example above, is known in Smartling as a Fuzzy Match Profile. You can create Fuzzy Match Profiles for all languages in your account, specific languages in your account, or specific languages in a workflow. You must create a Fuzzy Match Profile before associating it with a language or workflow.

Your Smartling Customer Success Manager (CSM) can help configure new Fuzzy Match Profiles for your projects if needed.

The Fuzzy Match Profile that was active when a string was translated is included in the Word Count report, along with the string’s fuzzy tier (shown as Fuzzy Breakdown in the report).

By default, fuzzy discounts apply only to the translation step, and editing is charged at the full per-word rate. Account Owners can also apply a Fuzzy Match Profile to the Post-Edit step or to all post-translation steps in Account Settings > Fuzzy Match Settings. In Smartling Language Services workflows, the profile always applies to the Human Translation and Post-Edit SLS steps.

The Fuzzy Breakdown column in the Word Count report displays either the fuzzy tier for available TM matches or the Language Quality Estimation level (Low, Medium, or High) if the translated string underwent Language Quality Estimation. See Language Quality Estimation Agent for Machine Translation for more details.

Weighted words simplify price calculations and help reduce the chance of errors. Instead of multiplying the number of source words in each fuzzy tier by the tier’s pricing percentage and the per-word price, you can simply multiply the weighted word count by the full per-word price. The discount is already factored into the weighted word count.

For example, consider the first row of the sample Word Count report in What are word counts and how do I read the Word Count report?. It contains 120 source words, all in the 95%–99.9% tier, which the example Fuzzy Match Profile charges at 30% of the full rate. Assuming a per-word price of $0.10:

The following table shows how the same 120 source words are priced in each tier of the example Fuzzy Match Profile, assuming a per-word price of $0.10.

In the sample report, the Edit row has no Fuzzy Match Profile, so its weighted words equal its word count (120).

Weighted words are not always source words with a fuzzy discount applied. In steps where no fuzzies are applied, such as MT, AI Translation and Light Post-Edit steps, there is no Fuzzy Match Profile discount. In MT and AI Translation steps, if SmartMatches are the only other line item, the weighted word count is the source word count minus the SmartMatches. See Machine or AI Translation steps.

Desktop Publishing (DTP) usually incurs an additional cost, invoiced as a separate line item based on the number of pages that require DTP. To make budgeting easier, Smartling lets you include the estimated DTP cost directly in your cost estimates.

The following table lists the main rules for who is credited for work in the Word Count report in each situation:

The additional translation work required for strings ingested with plural forms is reflected in the Word Count report. Plural forms are counted using a locale-specific multiplier. The number of source words is multiplied by the number of plural forms for each target locale.

A cost estimate predicts a job’s final word count and cost based on the content, Fuzzy Match Profiles, and rate cards in effect when the estimate is generated. It provides an approximation of how large the job is and what it would cost to translate at that point in time.

The following screenshot shows an example of a cost estimate in the Smartling dashboard.

Cost estimates are available only to users with the Account Owner, Project Manager, or Requester role. You can access them in the Job Summary pane of a job.

In addition to the estimate for the entire job available in the Job Summary, Smartling automatically runs a high-level estimate when you authorize a job. This estimate includes only unauthorized content, so it may differ from the total job estimate if some content has already been authorized. It displays only the top-level totals for source words, weighted words, and cost.

You can compare estimates for different workflows by selecting them from the dropdown in the Authorization dialog. To select separate workflows for each language, use the Show details option.

Agencies and translators can run fuzzy estimates for all workflow steps they have access to once content reaches those steps. Fuzzy tiers are applied only if a Fuzzy Match Profile is associated with the workflow step. Otherwise, the estimate displays only the word count for that step. Published content is not included in estimates.

With one exception, cost estimates do not account for actual historical data from the job. For example, you can run an estimate on a completed job. If a user who was not assigned to the edit step moved content past that step, effectively removing the editing cost, the estimate would not account for this. Instead, it assumes that the content would go through editing if the job were translated at the time the estimate was run.

The one exception is the fuzzy score used for translated strings. For these strings, the estimate uses the fuzzy score available at the time of translation rather than the score available in the current leveraged translation memory.

Cost figures are included in estimates when the appropriate rate cards have been configured in Smartling. A valid rate card must include a rate corresponding to the workflow step and content type. At least one assigned resource in every step of every workflow included in the estimate must have a valid rate card.

If some rate cards are missing, Smartling displays cost figures wherever possible (for example, for certain languages or workflow steps) and shows a warning next to the overall cost estimate. If the relevant rate cards use multiple currencies, Smartling does not calculate a total and displays a Multiple currencies message instead.

As translation work on a job progresses, the translation memory (TM) is gradually populated with entries from the job itself. These entries can then become available as potential fuzzy matches for other strings in the same job. Fuzzy matches that come from the job itself are called internal matches and are included in the estimate in their corresponding fuzzy tiers.

Identical (100%) internal matches are called repetitions. These are listed separately in the estimate because a separate price is often negotiated for repetitions.

Repetitions and internal matches are displayed only for untranslated content.

Repetitions that have not yet been translated in the computer-assisted translation (CAT) Tool appear in the estimate, even after machine translation (MT). However, once a linguist saves a translation for a repetition in the CAT Tool, it no longer appears in the estimate if the estimate is regenerated.

You may notice that an estimate shows 100% matches that don’t appear when you open the job in the CAT Tool.

This is because estimates include internal matches, which are matches between segments within the same job. These matches become visible in the CAT Tool only as you translate and save segments.

Estimates assume that all possible internal matches and repetitions (100% matches) will be used through auto-propagation. In the CAT Tool, these matches appear after a segment is translated and auto-propagation is applied.

Keep the following points in mind:

Strings eligible for SmartMatch are listed separately in the estimate and counted as zero weighted words for the translation step because their translations are applied automatically. SmartMatch removes the translation effort, which is why strings that go straight to Published count as zero weighted words.

If strings are SmartMatched to Published, they are excluded from the counts for all post-translation steps. However, if strings are SmartMatched to a post-translation step other than Published, the estimate assumes they will go to the first post-translation step and includes them in the counts for all post-translation steps. As a result, the estimate may overstate the amount of work required for this content.

For accounts without an AI Subscription, no costs are estimated for machine translation (MT) steps where the translation provider is an MT engine or a large language model (LLM), because no human linguists are involved. For accounts with an active AI Subscription, the estimate shows a cost for these steps at the AI Hub rate. In both cases, the fuzzy breakdown may still be visible for these steps, but it is not used in cost calculations.

No fuzzy discounts are applied to MT or AI Translation steps. Their weighted words are the source word count minus any SmartMatches.

Cost estimates are calculated for the AI Translation step in Smartling Language Services AIT workflows, and for the Light Post-Edit (LPE) step, which applies to group 2 locales.
​

Fuzzy discounts are never applied to either of these steps, including the human step (LPE), regardless of any translation memory (TM) matches. Because there is no Fuzzy Match Profile discount, the weighted word count in the AI Translation step is the source word count minus any SmartMatches, which are listed separately in the estimate. For example, if 100 source words have a 100% TM match and none are SmartMatched, the AI Translation step counts 100 weighted words. If 20 of those words are SmartMatched, it counts 80.
​

The LPE step is included in the AIT workflow rate, so it can show no rate and a cost of $0.00 in the estimate. Its cost is already part of the AIT rate.
​

SmartMatches are handled separately from fuzzy matches. Where a SmartMatch goes depends on your SmartMatch settings, which can differ for each match type:

In Smartling Language Services AI-Powered Human Translation (AIHT) workflows, fuzzy discounts are not used in the cost of the AI Translation step. The estimate may still show the fuzzy breakdown and discounted weighted words for this step, but they are not used in cost calculations. The AI Review step does have fuzzy discounts. They are based on the fuzzy tiers in the Fuzzy Match Profile associated with the step.

If your workflow includes a Smartling Language Services (SLS) managed LQA step, its cost is included in your cost estimate. This step is billed per word and uses the rate in your SLS rate card.

Fuzzy match discounts do not apply to the LQA step. Its cost is calculated on the full source word count.

Plural forms are not reflected in cost estimates. Instead, each string is counted as a single source string with one corresponding translation. In the Word Count report, however, plural forms are counted, and a multiplier is applied to the source word count.

Cost estimates are calculated using the following steps:

For each string in the job, identify a workflow for each target language to determine which workflow steps to include and which rate cards to use.

If a string already has a translation, the estimate uses the fuzzy score that was available when the string was translated. This score does not change when you refresh the estimate, even if the translation memory (TM) has changed since then. Refreshing an estimate picks up new TM matches only for strings that have not yet been translated.

For example, a string is translated when its best TM match is 85%. Later, the TM gains a 100% match for that string. After you refresh the estimate, the string still uses the 85% score. A string that is not yet translated would use the new 100% match.

For untranslated strings, the following checks are performed to determine the applicable fuzzy tier and calculate weighted word counts:

Weighted words are not always source words with a fuzzy discount applied. In MT and AI Translation steps, where no fuzzies are applied, the weighted word count is the source word count minus the SmartMatches. For example, if a step has 100 source words and 20 are SmartMatched, the weighted word count is 80. See Machine or AI Translation steps.

For each workflow step, use the rate card corresponding to the workflow’s content type and the rate corresponding to the step type to calculate the cost. If multiple rate cards with different rates could apply, the estimate displays a cost range.

Calculations are performed to four decimal places and rounded up to two decimal places at the end.

Add together the counts for all strings, workflow steps, and languages. The totals include:

Estimates that are re-run at different times over the life of a job can produce significantly different results. This is because an estimate is a real-time prediction of the job’s final word counts based on the current situation, which may have changed from when previous estimates were run. Some of the factors that contribute to estimates changing are:

With the exception of TM changes that occur as a result of normal translation work, the changes above can be managed to reduce their impact on active translation jobs.

Nonetheless, it’s advisable to save a copy of the original estimate, and the estimate when the content has entered its workflows, as well as any estimates run after significant changes of the types mentioned above.

Actual word counts and costs can differ from estimates for several reasons, including:

This section describes the known limitations of cost estimates and what you can do to account for each one.
​

An estimate reflects the state of a job when you run it, so running an estimate for the same job at different times can produce different results.

To manage this:

Once a translation is saved, subsequent estimates use the fuzzy score that was available at the time it was saved. Refreshing the estimate does not update the score for already translated strings, even if the TM has since changed. For an example, see Step 2: Calculate fuzzy scores and weighted words.
​

Estimates assume that all content destined for a dynamic workflow will follow its default branch. This applies even if some content is already in a different branch. As a result, dynamic workflows are not fully accounted for in estimates, which could lead to an underestimate.

Consider the following approaches when estimating dynamic workflows:

Estimates assume that content SmartMatched to a step other than Published will go to the first post-translation step. As a result, the estimated cost may be higher than the actual cost for this content. See How does SmartMatch affect estimates?

Strings longer than 10,000 characters are assumed to have no fuzzy matches. However, they are still included in the estimate’s total word count.
​

Estimates count a string with plural forms as a single source string. The Word Count report counts plural forms and applies a multiplier. See How are plural forms counted in cost estimates?
​

For accounts without an AI Subscription, machine translation (MT) steps are currently treated like other workflow steps and are therefore expected to have a rate card. Accounts with an active AI Subscription use subscription rates for these steps, so this limitation does not apply to them.

To avoid warnings about missing rates for these steps, consider creating a dummy user with an appropriate rate (for example, 0 or unpayable) and assigning that user to the MT steps.
​

Take warnings next to total cost figures seriously, as they can indicate that the total cost is significantly underestimated. Review the estimate details for all languages to identify which rates or costs are missing, and address any gaps before using the estimate.

Cost estimates are generally useful for budgeting because deviations in actual word counts tend to reduce the final cost compared with the estimated amount. However, the following exceptions can increase costs:

Ideally, you can avoid the first three types of changes once a job is in progress. If changes are unavoidable, you may need to run a new estimate and adjust the budget accordingly. For dynamic workflow considerations, see Dynamic workflows are estimated using the default branch.

One challenge when using estimates for budgeting is managing significant differences between estimated and actual costs. If the actual cost is higher than estimated, you or your translation vendor may need to cover the difference to ensure translators are paid for their work. Conversely, a lower-than-expected actual cost can create challenges when managing translator compensation.

Smartling filters out common stop words when indexing strings for search. These words are not stored in the search index and are therefore excluded from fuzzy matching in the computer-assisted translation (CAT) Tool and from leverage calculations in cost estimates.

If a string contains both stop words and non-stop words, only the non-stop words are indexed. If a string consists entirely of stop words, such as “no” or “not,” it will not generate fuzzy matches or contribute to leverage in estimates.

For English, the stop-word list includes: a, an, and, are, as, at, be, but, by, for, if, in, into, is, it, no, not, of, on, or, such, that, the, their, then, there, these, they, this, to, was, will, with.