English/German AI glossary
Which term will I hear in an English guide or a German lecture?
Use these brief definitions to connect the vocabulary, then follow a guide for a worked example.
In brief
- German lectures may use the English term, a translation or both.
- The definitions describe how the terms are used in these guides.
- A term such as agent or grounding does not establish what a particular application can do.
The German column gives useful counterparts rather than a fixed vocabulary for every lecturer. Acronyms such as RAG and DOI often remain the same in both languages. Follow the link in an English term to see how it applies to a study task.
Systems and generation
| English | German counterpart | Meaning in these guides |
|---|---|---|
| AI | Künstliche Intelligenz (KI) | Broad term for systems that perform tasks such as prediction, generation or decision support. |
| Generative AI | Generative KI | AI that produces new content such as text, images or audio. |
| Language model | Sprachmodell | Model that assigns probabilities to language sequences and can generate text. |
| Model | Modell | Trained computational system used to produce a prediction or other output. |
| Application | Anwendung | Product around a model, including its interface, instructions, data access and tools. |
| Agent | KI-Agent | System in which a model directs steps and tool use in a task loop. Product definitions vary. |
| Training | Training | Process that changes model parameters using data and an optimization objective. |
| Pre-training | Vortraining | Initial broad training that develops a model's general patterns and capabilities. |
| Post-training | Nachtraining | Further training that adapts capabilities or behavior after pre-training. |
| Inference | Inferenz / Modellausführung | Running a trained model to produce an output. |
Input, context and tools
| English | German counterpart | Meaning in these guides |
|---|---|---|
| Prompt | Prompt / Eingabe | Instructions or other input given to an AI system for a task. |
| System instruction | Systemanweisung | Instruction supplied by the application alongside user input. |
| Token | Token | Unit used by a model's tokenizer, such as a word part, punctuation or another piece of text. |
| Tokenization | Tokenisierung | Conversion of content into the tokens used by a model. |
| Context | Kontext | Information supplied to the model for a particular response. |
| Context window | Kontextfenster | Limited token capacity for an interaction, including applicable input and output. Accounting differs by model. |
| Tool call | Werkzeugaufruf | Request from a model-driven system to use a connected tool. |
| Permission | Berechtigung | Granted access or authority for a system to read data or take actions. |
| Prompt injection | Prompt-Injection / eingeschleuste Anweisung | Instruction hidden in material that tries to redirect a system from its task. |
| Local model | Lokal ausgeführtes Modell | Model whose computation runs on a device or a locally controlled machine. Other application features may still contact external services. |
Evidence and learning
| English | German counterpart | Meaning in these guides |
|---|---|---|
| Hallucination | Halluzination | Generated content that is false or unsupported, including invented references. |
| Grounding | Quellenbindung / Grounding | Supplying external material as a basis for an answer. It does not guarantee faithful use. |
| Retrieval | Informationsabruf / Retrieval | Searching a collection for material relevant to a question. |
| RAG | Retrieval-augmented generation (RAG) | Generation using retrieved material added to the response context. |
| Citation | Quellenangabe / Zitatnachweis | Reference identifying a source. Its existence, metadata and support for the claim still need checking. |
| DOI | Digital Object Identifier (DOI) | Persistent identifier for a registered work or other object. It does not certify the work's claims. |
| Verification | Überprüfung | Checking a claim or result against suitable evidence or tests. |
| Cognitive offloading | Kognitive Auslagerung | Transferring mental work to another person or tool. Whether it helps depends on the learning goal. |
| Retrieval practice | Abrufübung | Trying to recall information from memory as a learning activity. This differs from document retrieval. |
| Delegation | Delegation / Aufgabenübertragung | Giving a defined part of a task to another person or system. |
| Diff | Änderungsvergleich / Diff | Comparison showing what changed between two versions of a file. |
Evaluation
| English | German counterpart | Meaning in these guides |
|---|---|---|
| Bias | Verzerrung | Systematic difference in outputs or treatment that can disadvantage people or distort results. |
| Benchmark | Benchmark / Vergleichstest | Defined test used to compare systems under stated conditions. |
| Model card | Modellkarte | Document describing a model's intended use, evaluation and limitations. |
| System card | Systemkarte | Document describing a deployed system, including its components, evaluation and safeguards. |
Sources and scope
Terminology review: 5 October 2026. These definitions summarize the usage in the linked guides, whose primary sources and scope notes give the basis for more detail. German counterparts are editorial translations for this course, not an official terminology standard.
For technical examples, see tokenization in the official tiktoken repository, context in Anthropic's documentation and workflows and agents in Anthropic's engineering guide. Crossref explains DOI metadata checks. The retrieval-practice paper by Roediger and Karpicke supports the learning term, which should not be confused with searching a document collection.