TER

TER, or Translation Edit Rate, is a metric used to evaluate machine translation quality by measuring how many edits are required to change a machine-generated translation into a correct reference translation.

Instead of measuring similarity alone, TER focuses on the amount of effort needed to fix a translation. A lower TER score indicates that fewer edits are required and the translation is closer to the desired result.

How TER Works

TER evaluates translation output by calculating the edits required to match a reference translation.

Edit Operations The metric counts insertions, deletions, substitutions, and word shifts needed to correct the translation.

Reference Comparison The machine translation is compared against a human reference translation.

Score Calculation The number of edits is divided by the total number of words to generate the TER score.

Quality Measurement Lower TER scores generally indicate higher translation quality.

Why TER Matters

TER provides a practical way to estimate translation effort.

• Measures the editing effort required to correct translations
• Helps evaluate machine translation system performance
• Supports benchmarking between translation models
• Provides insight into post-editing workload

TER in Modern Translation Evaluation

TER is commonly used alongside other evaluation metrics, such as BLEU, to measure machine translation performance. While automated metrics help benchmark translation systems, many organizations also combine them with human review to assess translation quality.

LILT’s AI-powered translation platform uses advanced evaluation methods and continuous human feedback to improve translation performance and deliver high-quality multilingual content.