Can AI Translate Psychoeducational Reports Accurately? What Our New Study Found

Psychoeducational reports contain information that families need to understand. Yet these reports are often long, technical, written at a high reading level, and available only in English.

That creates a serious access problem for multilingual families. It also raises a practical question for school psychologists: Can tools such as Google Translate or ChatGPT help translate psychoeducational reports accurately?

A new article that I coauthored in Contemporary School Psychology examines this question. The study, “Lost in Translation? Evaluating Machine Translation for Psychoeducational Reports,” was written by Gagan Shergill, Scott Stage, Adam Lockwood, Laura Rodriguez, and Crucita Adams.

The article is open access, which means the full paper is free for everyone to read and download.

The short answer is that machine translation showed promise, but no translation method was error-free. Human review, professional judgment, privacy protections, and clearer report writing remain essential.

Key Points for Readers on the Go

- The study compared English-to-Spanish translations produced by Google Translate, ChatGPT-4o, and a professional human translator.

- The three approaches did not differ significantly in fluency.

- Overall accuracy differed across the three approaches, but corrected follow-up comparisons did not establish a statistically significant advantage for any specific pair.

- The total number of observed errors differed across the translation approaches.

- Errors affecting meaning occurred in machine-generated and human translations.

- The findings are preliminary and should not be interpreted as evidence that one translation method is definitively superior.

- A human-in-the-loop workflow may be useful, but school psychologists remain responsible for accuracy, confidentiality, and appropriate professional use.

- The complete article is open access and free for everyone to read.

What We Studied

The study used a fictional 71-sentence psychoeducational report summary. The report included common assessment information related to cognitive abilities, phonological processing, academic achievement, and the evaluation of a possible specific learning disability.

The English report was translated into Spanish using three approaches:

1. Google Translate

2. ChatGPT-4o

3. A professional Spanish-English human translator

Two bilingual graduate students evaluated the translations for fluency, accuracy, and specific error types. Disagreements were resolved by a bilingual school psychology doctoral student who was blinded to the source of each translation.

This design allowed the research team to examine more than whether a translation sounded natural. The evaluators also considered whether it preserved the meaning of the original psychoeducational information.

The Translations Often Sounded Fluent

The study did not find significant differences in fluency across Google Translate, ChatGPT-4o, and the human translation.

That finding matters because fluent language can create a false sense of confidence. A translated sentence may sound polished and grammatically acceptable while still changing or obscuring the intended meaning.

For psychoeducational reports, sounding right is not enough. The translation must accurately communicate technical concepts, student strengths and needs, assessment findings, and educational recommendations.

No Method Produced a Perfect Translation

The study found a statistically significant overall difference in accuracy across the three translation approaches. However, after correcting for multiple comparisons, the follow-up tests did not show a statistically significant difference between any specific pair of approaches.

The total number of observed errors also differed across the systems. In this particular study, the human translation had the highest cumulative number of errors, followed by Google Translate and ChatGPT-4o.

That finding should be interpreted cautiously. The study involved one fictional report, one human translator, one language pair, and a relatively small number of sentences. It does not establish that machine translation is generally more accurate than professional human translation.

Instead, it highlights a more useful point: the label attached to a translation method does not guarantee accuracy. The quality and domain-specific preparation of a human translator matter. The configuration and use of a machine translation system matter. In every case, the output requires review.

The Errors Could Affect Interpretation

Errors appeared across all three approaches. These included mistranslations, grammatical problems, untranslated words, and critical inaccuracies that altered meaning.

Technical terms were especially challenging. Examples included concepts related to phonological processing, rapid automatic naming, comprehension knowledge, and assessment terminology.

Some errors were subtle. A translation could preserve the general topic while changing the precision, formality, subject, or intended meaning of a sentence. In a high-stakes document, even a subtle change may affect how a family understands a student’s evaluation.

Machine Translation Should Not Be a Copy-and-Send Workflow

The study does not support pasting a confidential report into a public tool, accepting the first translation, and sending it to a family.

A more defensible approach is a human-in-the-loop workflow. A machine translation tool may produce an initial draft, but a qualified person with relevant linguistic and psychoeducational expertise should review and refine it.

School psychologists must also determine whether a translation platform provides the protections required for their setting. Personally identifiable student information should not be entered into a tool unless its data handling, access controls, and privacy protections are appropriate. Depending on the setting, FERPA, HIPAA, professional ethics, and local policies may apply.

Technology may assist with translation. It does not transfer professional responsibility away from the practitioner.

Better Translation Begins with Better Report Writing

One of the most practical implications is that the original English report affects the quality of every subsequent translation.

Reports are easier to translate when they:

- use clear and concise sentences;

- avoid unnecessary jargon and idiomatic language;

- define specialized terms;

- use active voice when appropriate;

- connect findings directly to referral questions and recommendations;

- maintain consistent terminology; and

- are written for families rather than only for other professionals.

A readable report is not only easier to translate. It is also more useful to English-speaking families, educators, and other members of the educational team.

What School Psychologists Can Do Now

The evidence is still preliminary, but several cautious practices are reasonable:

1. Do not assume fluency equals accuracy. Review whether the translation preserves the meaning of the source report.

2. Use appropriate privacy protections. Confirm that the selected platform and workflow are suitable for confidential educational or clinical information.

3. Include qualified human review. Whenever possible, involve someone with linguistic expertise and familiarity with psychoeducational terminology.

4. Give translators context and resources. Glossaries, intended-audience information, and opportunities to clarify ambiguous language may improve quality.

5. Write the source report clearly. Reducing jargon and unnecessary complexity can improve both accessibility and translatability.

6. Document professional oversight. The school psychologist remains accountable for the accuracy and appropriate use of the final report.

Important Limitations

This was an exploratory study. It examined one fictional report, one professional human translator, one LLM, one dedicated machine translation platform, and English-to-Spanish translation without specifying a regional dialect.

The study also did not evaluate every potentially important dimension of translation quality, such as tone, stylistic consistency, cultural responsiveness, or performance across different Spanish dialects.

The findings therefore raise useful questions and identify promising practices, but they should not be treated as a final comparison of human and machine translation.

Bottom Line

Machine translation may eventually help schools address persistent language-access barriers. In this study, Google Translate and ChatGPT-4o produced translations that were often fluent and reasonably accurate. However, neither machine translation nor human translation was perfect.

The practical lesson is not to choose technology over people, or people over technology. It is to build stronger translation workflows.

For school psychologists, that means using appropriate privacy protections, improving the readability of source reports, preparing translators to work with psychoeducational terminology, and ensuring meaningful human review before translated information reaches families.

The article is open access and free for everyone to read and download:

https://doi.org/10.1007/s40688-026-00608-4

AI Disclosure: Generative AI was used to assist with drafting and editing this post and to create the accompanying image. I reviewed and revised the content and take responsibility for its accuracy and final form.