To translate employee performance reviews, appraisals and HR feedback into any language, the target must preserve the manager’s evaluation rather than merely reproduce the words. People searching for performance review translation, appraisal translation, HR feedback translation, employee evaluation translation or AI translation of review documents need natural target language without changing rating strength, evidence, praise, concerns, development goals, behavioural expectations, promotion-readiness signals or the distinction between observation, judgment and recommendation.
Word-for-word translation can still distort an appraisal because HR language is calibrated. “Meets expectations,” “exceeds expectations,” “developing,” “inconsistent,” “needs support,” “ready for broader responsibility,” “not yet ready,” “demonstrates potential,” and “requires improvement” occupy different positions on an evaluation scale. A fluent target can become unfair if it strengthens mild developmental feedback into criticism, softens a formal performance concern, or turns a manager’s recommendation into a guaranteed outcome.
This guide develops a practical method for translating employee performance reviews, appraisals and HR feedback without changing evaluation or tone. It covers rating scales, goals, evidence, strengths, development areas, behaviours, outcomes, calibration, manager comments, employee comments, promotion and compensation signals, improvement expectations, confidentiality, AI and machine translation, worked examples, practice and final quality assurance.
The Core Performance-Review Translation Principle
Translate the evaluation state exactly: evidence → rating → interpretation → strength → development need → recommendation → next review action.
A performance review is both record and judgment. Translation should keep the evidence separate from the evaluator’s interpretation, and keep both separate from future recommendations. The target should neither reward nor penalize the employee through accidental wording changes.
The Ten-Part Translation Method
- 1. Rating Scale: preserve each formal rating and its position on the scale.
- 2. Evidence vs Evaluation: keep observed facts distinct from the manager’s judgment.
- 3. Strength of Praise: preserve positive evaluation without inflation.
- 4. Developmental Feedback: keep improvement language specific and constructive.
- 5. Behaviour vs Outcome: preserve whether the review judges how work was done or what result was achieved.
- 6. Goal Status: keep completed, partially completed, deferred and missed goals distinct.
- 7. Calibration Language: preserve comparative or organizational context without inventing rank.
- 8. Promotion and Readiness Signals: keep recommendation strength and timing exact.
- 9. Performance Concerns: preserve seriousness, recurrence and formal status.
- 10. Manager and Employee Voice: keep attribution and disagreement visible.
1. Rating Scale
A common failure point is mapping unfamiliar ratings onto target words with stronger or weaker force. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is building a glossary from the organization’s own rating definitions before translating comments. This gives the translator a stable map before target-language editing begins.
Worked example: “Meets expectations” may be a solid successful rating, not faint praise and not the same as “average.” The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to reconstruct the rating ladder from the target and compare it with the source. If the result changes, the translation has altered the document rather than merely the wording.
Scale control helps school reports, surveys and quality audits. Reusing the same structured check turns one correction into a transferable translation skill.
2. Evidence vs Evaluation
A common failure point is turning examples into general traits or judgments into facts. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is marking each sentence as observation, result, interpretation or recommendation. This gives the translator a stable map before target-language editing begins.
Worked example: “Delivered three projects on schedule” is evidence; “demonstrates reliable planning” is an interpretation built from evidence. The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to trace every target judgment back to the same source evidence. If the result changes, the translation has altered the document rather than merely the wording.
Evidence discipline transfers to incident reports and references. Reusing the same structured check turns one correction into a transferable translation skill.
3. Strength of Praise
A common failure point is choosing more enthusiastic adjectives because they sound professional in the target language. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is ranking praise terms before selecting equivalents. This gives the translator a stable map before target-language editing begins.
Worked example: “Consistently reliable” should not automatically become “exceptional” or “outstanding.” The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to rate source and target praise intensity independently. If the result changes, the translation has altered the document rather than merely the wording.
Intensity mapping helps testimonials and recommendation letters. Reusing the same structured check turns one correction into a transferable translation skill.
4. Developmental Feedback
A common failure point is turning a narrow development need into a broad negative label. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is separating behaviour, impact and desired change. This gives the translator a stable map before target-language editing begins.
Worked example: “Needs to delegate earlier on complex projects” is not the same as “poor leadership.” The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to state the exact improvable behaviour from the target. If the result changes, the translation has altered the document rather than merely the wording.
Developmental language control helps education and coaching. Reusing the same structured check turns one correction into a transferable translation skill.
5. Behaviour vs Outcome
A common failure point is merging behavioural competencies with output metrics. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is mapping each statement to behaviour, outcome or both. This gives the translator a stable map before target-language editing begins.
Worked example: A salesperson may exceed revenue targets while still needing improvement in CRM documentation. The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to build separate behaviour and outcome summaries from the target. If the result changes, the translation has altered the document rather than merely the wording.
Dimension separation supports scorecards and competency frameworks. Reusing the same structured check turns one correction into a transferable translation skill.
6. Goal Status
A common failure point is using generic success or failure language for mixed progress. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is mapping goal, target, deadline, evidence and status before translation. This gives the translator a stable map before target-language editing begins.
Worked example: A goal completed late is not the same as a goal completed on time or not completed. The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to reconstruct goal status from target only. If the result changes, the translation has altered the document rather than merely the wording.
Status mapping helps project and audit documents. Reusing the same structured check turns one correction into a transferable translation skill.
7. Calibration Language
A common failure point is turning relative wording into numerical ranking or vice versa. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is keeping comparison group, time period and confidence level explicit. This gives the translator a stable map before target-language editing begins.
Worked example: “One of the stronger contributors in the team” is not “top 10%” unless the source states that. The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to compare source and target comparison scope. If the result changes, the translation has altered the document rather than merely the wording.
Comparison control helps academic and corporate reporting. Reusing the same structured check turns one correction into a transferable translation skill.
8. Promotion and Readiness Signals
A common failure point is turning potential or readiness discussion into a promotion promise. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is separating current capability, potential, recommendation and decision authority. This gives the translator a stable map before target-language editing begins.
Worked example: “Ready to be considered for broader scope” is not “will be promoted.” The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to state what is and is not promised from the target. If the result changes, the translation has altered the document rather than merely the wording.
Recommendation-state control helps hiring and succession planning. Reusing the same structured check turns one correction into a transferable translation skill.
9. Performance Concerns
A common failure point is softening a formal concern into optional coaching or exaggerating a minor issue. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is mapping issue, evidence, frequency, prior feedback and expected correction. This gives the translator a stable map before target-language editing begins.
Worked example: “Repeated missed deadlines after prior coaching” carries more seriousness than “could improve time management.” The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to rank source and target concern severity. If the result changes, the translation has altered the document rather than merely the wording.
Severity control supports complaints and disciplinary records. Reusing the same structured check turns one correction into a transferable translation skill.
10. Manager and Employee Voice
A common failure point is blending manager comments and employee response into one consensus narrative. Because this document type mixes factual record, evaluation and action, a fluent target can still shift meaning if the translator does not separate those layers first.
The mechanism is tagging who says each evaluative statement and preserving disagreement where it exists. This gives the translator a stable map before target-language editing begins.
Worked example: An employee may disagree with a rating while acknowledging the underlying event. The acceptance test is whether a target-language reader would reach the same conclusion, responsibility or next action as a source-language reader.
A reliable check is to reconstruct each party’s position separately from the target. If the result changes, the translation has altered the document rather than merely the wording.
Attribution discipline supports meetings, grievances and legal records. Reusing the same structured check turns one correction into a transferable translation skill.
Worked Example Laboratory
Example 1: Strong Result, Moderate Behaviour Rating
Employee exceeded revenue goal but received “meets expectations” for collaboration. Two dimensions differ and should remain separate.
Do not let the strong numeric result inflate the behavioural rating. The goal is to preserve the source record, not to improve or reinterpret it.
Example 2: Developmental Recommendation
“Would benefit from earlier escalation of risks.” The comment suggests a change in timing, not a broad lack of judgment.
Translate the specific behaviour and advisory force. The goal is to preserve the source record, not to improve or reinterpret it.
Example 3: Promotion Readiness
“Demonstrates potential for a team-lead role with further coaching.” Potential and development requirement coexist.
Keep both; do not convert potential into confirmed readiness. The goal is to preserve the source record, not to improve or reinterpret it.
Example 4: Formal Concern
“Accuracy remains below the required standard despite prior feedback.” The statement contains threshold, persistence and prior intervention.
Preserve seriousness without adding disciplinary consequences not stated. The goal is to preserve the source record, not to improve or reinterpret it.
Example 5: Employee Response
“I agree with the project feedback but disagree with the communication rating.” The employee accepts one point and contests another.
Keep the split response rather than summarizing it as agreement or disagreement overall. The goal is to preserve the source record, not to improve or reinterpret it.
Rating Systems Are Local to the Organization
Do not assume that familiar labels carry universal meaning. Some organizations use “3” as a successful rating and “5” as exceptional; others use different language or forced distributions. Translate according to the source organization’s documented definitions where available.
If readers need context, preserve the source rating label and provide an explanation rather than inventing cross-company equivalence.
Calibration and Bias
Translation should not add demographic, cultural or personality assumptions that are absent from the source. Keep comments tied to observable work behaviour, outcomes and documented competencies.
If a phrase such as “executive presence” or “culture fit” is institution-specific or vague, seek context rather than choosing a target term that adds unsupported traits.
Development Plans and Improvement Expectations
Development plans may be supportive, while formal performance-improvement processes may carry stronger procedural meaning. Preserve the document’s status and do not make a coaching plan sound disciplinary unless the source does.
Likewise, do not soften a formal requirement into optional professional development.
Confidentiality and Sensitive HR Data
Performance reviews contain personal and employment information. Translation workflows should use authorized systems and minimize unnecessary exposure. Where AI or machine translation is used, follow the organization’s confidentiality and data-handling requirements.
Privacy controls are part of the workflow, but they should not alter the content of the review itself.
Whole-Review Consistency
Read the entire appraisal after translation. A paragraph-by-paragraph translation can accidentally create contradictions: “consistently exceeds expectations” in one section and a weaker summary rating in another. The target should preserve the same internal logic as the source.
Practice and Checking
Practice 1: Rating Ladder
Translate five rating labels and definitions as a set. Complete the first pass manually so the document logic remains visible before tool assistance.
Ask target-language reviewers to order them by strength and compare with the source scale. Record errors by category so repeated weaknesses become a targeted practice plan.
Practice 2: Evidence Trace
Highlight every concrete example in a review. Complete the first pass manually so the document logic remains visible before tool assistance.
Link each target evaluation statement to the same evidence. Record errors by category so repeated weaknesses become a targeted practice plan.
Practice 3: Praise Calibration
Translate several positive comments of different strength. Complete the first pass manually so the document logic remains visible before tool assistance.
Rank source and target intensity independently. Record errors by category so repeated weaknesses become a targeted practice plan.
Practice 4: Development Rewrite Test
Translate narrow improvement comments. Complete the first pass manually so the document logic remains visible before tool assistance.
Check that none becomes a global personality judgment. Record errors by category so repeated weaknesses become a targeted practice plan.
Practice 5: Promotion-State Test
Translate potential, readiness, nomination and decision language. Complete the first pass manually so the document logic remains visible before tool assistance.
State exactly what outcome is and is not guaranteed. Record errors by category so repeated weaknesses become a targeted practice plan.
Practice 6: Two-Voice Review
Translate manager and employee comments that disagree. Complete the first pass manually so the document logic remains visible before tool assistance.
Summarize each position separately from the target. Record errors by category so repeated weaknesses become a targeted practice plan.
Independent-Use Workflow
- Identify the organization’s rating scale and competency glossary.
- Lock formal ratings, goals, dates and quantitative results.
- Separate evidence, evaluation, development need and recommendation.
- Preserve praise and concern at the same intensity.
- Keep behavioural and outcome dimensions distinct.
- Track goal status and deadline performance accurately.
- Preserve promotion, readiness and compensation signals without inventing decisions.
- Keep manager and employee attribution separate.
- Run a whole-review consistency check.
- Perform final confidentiality and target-only interpretation QA.
Using AI and Machine Translation
AI can accelerate translation of recurring appraisal language when given the organization’s rating definitions, competency glossary and role context. Instruct it not to strengthen praise, soften formal concerns or infer promotion decisions.
A strong QA prompt asks the model to extract rating, evidence, strength, development area, goal status and recommendation from source and target separately. Compare the structured outputs and verify every mismatch manually.
Useful Internal Routing
For general translation reasoning, use The Universal Five-Layer Translation Method. For recommendation strength, see How to Translate Recommendation Letters, References and Testimonials Without Changing the Writer’s Evaluation.
For employment-entry language, use How to Translate Job Offers, Employment Contracts and Onboarding Forms Without Changing Pay, Probation or Employment Terms. For final QA, use How to Check Translation Accuracy Before You Send, Submit or Publish.
Frequently Asked Questions
Should performance reviews be translated literally?
No. They should be natural in the target language while preserving the exact evaluation scale, evidence, strength, concerns and recommendations.
Can AI translate appraisals?
Yes, especially with rating definitions and competency context, but praise intensity, formal concerns, promotion signals and attribution need careful review.
How do I translate “meets expectations”?
Use the organization’s defined meaning. Do not assume it is weak praise or average performance without context.
What is the risk with developmental feedback?
Turning a specific behaviour to improve into a broad negative trait or, in the other direction, softening a formal concern into optional advice.
How should promotion-readiness comments be handled?
Preserve whether the source states potential, readiness, recommendation, nomination or a final decision. These are different states.
Should employee comments be merged with manager comments?
No. Keep attribution visible, especially where the employee disagrees with a rating or interpretation.
How do I check the translated rating tone?
Compare source and target on a simple intensity scale and review the whole document for internal consistency.
What is the best final test?
Using only the target, state the employee’s ratings, strongest evidence, main development need, goal status and any future recommendation. Compare with the source.
The Rule to Keep
Performance-review translation is successful when the employee is evaluated at the same level for the same evidence, with the same praise, concerns and development expectations.
Translate the appraisal so language neither promotes nor penalizes the employee.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.
Deep Practice: Reconstruct the Performance Profile
Take the translated review and create a table with rating, evidence, outcome, behaviour, praise strength, development need, goal status and future recommendation. Build the same table from the source. Differences reveal evaluative drift faster than ordinary proofreading.
Then ask an independent reviewer to read only the target and describe the employee’s overall standing. If their interpretation is substantially stronger or weaker than the source reviewer’s interpretation, inspect adjectives, modal verbs, rating labels and recommendation language.