This post presents a structured application of the Gemini 2.5 Pro language model for the bilingual summarisation of selected Hungarian Constitutional Court decisions. Working exclusively from an official PDF file containing full-text rulings, the model was prompted to generate parallel summaries in Hungarian and UK English, along with a list of the constitutional and statutory provisions interpreted in each decision. The task was designed to simulate a legal research scenario requiring precision, domain knowledge, and multilingual output, demonstrating how GenAI can assist in processing complex legal texts for scholarly purposes.
Input file
The official PDF publication "A Magyar Köztársaság Alkotmánybíróságának határozatai 1991 (1–13. szám)", which compiles the Constitutional Court’s rulings from January to mid-April 1991, was the input for this task. The document is a scanned, text-recognised file containing full-text decisions in Hungarian, complete with reference numbers, constitutional reasoning, and citations of relevant legal provisions. The prompt positioned the model as a Hungarian-speaking legal analyst with constitutional expertise, capable of generating bilingual legal summaries. Each output consisted of a Hungarian and a UK English summary (150–250 words each), followed by a list—in Hungarian—of all laws and constitutional provisions reviewed in the ruling. This design prioritised legal precision, clarity, and terminological consistency across both languages.
Prompt
The model was tasked with producing structured summaries specific Hungarian Constitutional Court decisions from 1991, based solely on the content of the uploaded official PDF. Each decision was identified by its precise reference number, and the model was instructed to locate and interpret the relevant section without relying on external sources or hallucinated content.
Task
Summarise several Hungarian Constitutional Court decisions based only on the uploaded PDF.
Role / Persona
You are a Hungarian-speaking legal analyst with expertise in constitutional law. Your task is to locate each specified decision in the PDF and write a legally precise and accessible summary for each.
You must write the summary in both Hungarian and UK English.
Input
The uploaded PDF contains official Hungarian Constitutional Court decisions from the year 1991.
You are provided with a list of decisions to summarise, each identified by its exact reference number.
Target decisions
- 3/1991. (II. 7.) AB
- 7/1991. (II. 28.) AB
- 16/1991. (IV. 20.) AB
- 28/1991. (VI. 3.) AB
- 46/1991. (IX. 10.) AB
- 64/1991. (XII. 17.) AB
Instructions
- Locate the decision in the uploaded PDF using its exact reference number.
- Work only with the section that clearly belongs to that decision.
- Do not hallucinate or invent content. If the decision cannot be found in full, say so.
- Do not use the internet or external content. Work only with the uploaded file.
- Write a Hungarian-language summary using precise but accessible legal language (150–250 words).
- Then, write a UK English-language summary (150–250 words), maintaining legal accuracy and clarity.
- After both summaries, provide a separate Hungarian list of all laws, statutes, or constitutional provisions that the Court reviewed or interpreted in the decision.
Output format
Repeat this structure for each decision:
[e.g. 3/1991. (II. 7.) AB]
Összefoglaló (magyarul):
[Hungarian summary – 150–250 words]
Summary (in UK English):
[UK English summary – 150–250 words]
Az Alkotmánybíróság által felülvizsgált jogszabályok:
[List of laws, constitutional provisions, etc. – in Hungarian]
Output
The model produced consistently high-quality outputs across all targeted decisions contained in the official PDF. Although the rulings were not presented sequentially in the document and required careful identification based on reference numbers, Gemini 2.5 Pro successfully located each one without error. It avoided hallucinations, respected structural instructions, and delivered legally precise summaries in both Hungarian and UK English. The outputs reflected a strong command of constitutional legal language, maintaining terminological clarity and stylistic appropriateness in both languages. The Hungarian summaries were legally sound yet accessible, while the English versions adhered to the conventions of UK legal writing. The separate lists of reviewed legal provisions confirmed the model’s ability to extract structured information from complex, scanned judicial texts.
Recommendations
This experiment shows that Gemini 2.5 Pro can effectively support legal researchers in processing scanned court materials with high accuracy and structural discipline. The model handled bilingual summarisation and legal terminology with notable reliability, even when rulings were dispersed and embedded deep within the source document. Its capacity to follow strict prompts and avoid hallucinations makes it a practical tool for preparing first-pass analyses or enhancing access to archival legal texts.
The authors used Gemini 2.5 Pro Preview [Google DeepMind (2025) Gemini 2.5 Pro Preview (accessed on 20 June 2025), Large language model (LLM), available at: https://deepmind.google/technologies/gemini/] to generate the output.
