---
license: mit
language:
- en
tags:
- history
- holocaust
- oral testimonies
pretty_name: USHMM English Oral Testimonies Dataset
---
# Dataset Card for USHMM English Oral Testimonies Dataset

## Dataset Description

- **Homepage:** https://www.ushmm.org/collections/the-museums-collections/about/oral-history

### Dataset Summary

This is a collection of approximately 1,000 English Oral Testimonies at the United States Holocaust Memorial Museum (USHMM). The oral testimonies were collected during the late-twentieth and early twenty-first centuries. These were converted from PDFs into raw text with [Tesseract](https://github.com/tesseract-ocr/tesseract). The text was post-processed with a Python script to convert it into segments of dialogue. Because this process was automated, mistakes may remain. Occasionally, headers and footers appear in the middle of the dialogue. If found, submit an issue and these can be corrected.

This dataset was created during William J.B. Mattingly's postdoc at the Smithsonian Institution's Data Science Lab which had a cross-appointment with the USHMM. This dataset is being used for text classification, named entity recognition, and span categorization.

### Languages

These testimonies are strictly in English, but they were given by non-native speakers. This means foreign language words and phrases may appear throughout the testimonies.

## Dataset Structure

### Data Fields

- **rg:** String, the RG number used by the USHMM to identify specific items in a collection.
- **sequence:** Integer, the unique ID for the dialogue row.
- **text:** String, the actual piece of dialogue.
- **category:** String, can be a question or an answer.

### Data Splits

The dataset is not split into train, test, or validation sets.

## Dataset Creation

### Curation Rationale

The dataset was created to make the testimonies more accessible for various machine learning tasks. It is also the first publicly available dataset for Holocaust oral testimonies.

### Source Data

#### Initial Data Collection and Normalization

The initial data was collected from the United States Holocaust Memorial Museum's (USHMM) Oral Testimonies. These testimonies were converted from PDFs into raw text with Tesseract and then post-processed with a Python script to convert them into segments of dialogue.

#### Who are the source language producers?

The source language producers are the survivors of the Holocaust who shared their experiences during the Oral Testimonies collected by the USHMM.

### Personal and Sensitive Information

The dataset contains personal narratives and testimonies of Holocaust survivors which may include sensitive information.

## Considerations for Using the Data

### Social Impact of Dataset

This dataset provides invaluable insights into the experiences of Holocaust survivors. It can aid in historical studies, and also serve as a rich resource for Natural Language Processing tasks related to understanding dialogues, emotion, sentiment, and other semantic and syntactic features of language.

### Discussion of Biases

As the dataset is based on personal testimonies, it is subjective and can contain the personal biases of the people sharing their experiences.

### Other Known Limitations

Since the testimonies were converted from PDFs into raw text using Tesseract, there may be OCR errors. Also, as the testimonies were given by non-native English speakers, there can be instances of imprecise English and foreign language words or phrases.

## Additional Information

### Dataset Curators

The dataset was curated by [William J.B. Mattingly](https://github.com/wjbmattingly).

### Licensing Information

Forthcoming

### Citation Information

USHMM Oral Testimonies Dataset. Curated by William J.B. Mattingly.

### Contributions

If you wish to contribute, please feel free to submit an issue.