tab / README.md
NathanRoll's picture
init dataset card
ebc20fd verified
---
dataset_info:
features:
- name: question
dtype: string
- name: subtest
dtype: string
splits:
- name: train
num_bytes: 1463
num_examples: 20
download_size: 2343
dataset_size: 1463
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: mit
task_categories:
- text-generation
- text-classification
language:
- en
tags:
- aphasia
pretty_name: Text Aphasia Battery (TAB)
size_categories:
- n<1K
---
# Text Aphasia Battery (TAB)
The **Text Aphasia Battery (TAB)** is a modified subset of the [Quick Aphasia Battery (QAB)](https://aphasialab.org/qab/) designed to assess aphasic symptoms in environments where input and output modalities are restricted to text. TAB is especially useful for large-scale identification of aphasic features in applications such as corpus studies and evaluations of large language models (LLMs). **Note:** TAB is not a replacement for the QAB or other clinical diagnostic tools.
---
## Overview
TAB consists of **four subtests** that evaluate different aspects of language function:
1. **Connected Text**
2. **Word Comprehension**
3. **Sentence Comprehension**
4. **Repetition**
Each subtest comes with specific instructions and scoring criteria, aimed at detecting various aphasic markers.
---
## Subtests
### 1. Connected Text
- **Objective:**
Evaluate fluency, grammaticality, and coherence of speech.
- **Instructions:**
Respond to the following prompts in 3–5 full sentences:
- "Tell me about the best trip you ever took."
- "Describe a happy childhood memory."
- "Tell me about your first job."
- "What do you like about where you live?"
- "Describe what is happening in the following scene: The boy is pushing the girl."
- **Scoring:**
Uses adapted non-motor APROCSA features (see **Notes on Evaluation**).
---
### 2. Word Comprehension
- **Objective:**
Evaluate lexical-semantic processing and selection among competing meanings.
- **Instructions:**
Elicit responses **verbatim** for the following:
- "Which one is an animal: ‘lion’ or ‘drum’?"
- "Which object is typically used to make music: ‘violin’ or ‘giraffe’?"
- "Which item is usually worn on the feet: ‘boot’ or ‘boat’?"
- "Which object is used for cutting: ‘knife’ or ‘kite’?"
- "Which one is a large mammal with a long neck: ‘giraffe’ or ‘horse’?"
- **Scoring:**
Responses are scored in a binary fashion (correct/incorrect). Incorrect selections may indicate deficits in semantic processing.
---
### 3. Sentence Comprehension
- **Objective:**
Evaluate syntactic processing, comprehension of passive structures, and logical reasoning.
- **Instructions:**
Elicit responses **verbatim** for the following:
- "Are babies watched by babysitters?" (Expected: **Yes**)
- "Do you cut the grass with an axe?" (Expected: **No**)
- "If you’re about to leave, have you left yet?" (Expected: **No**)
- "Are witnesses questioned by police?" (Expected: **Yes**)
- "If I was at the park when you arrived, did I get there first?" (Expected: **Yes**)
- **Scoring:**
Delayed or incorrect responses may indicate difficulties with syntactic processing, handling of negation, or confusion with passive constructions.
---
### 4. Repetition
- **Objective:**
Evaluate morphosyntactic integrity.
- **Instructions:**
Elicit responses **verbatim** for the following:
- "Please repeat exactly: house."
- "Please repeat exactly: breakfast."
- "Please repeat exactly: catastrophe."
- "Please repeat exactly: The sun rises in the East."
- "Please repeat exactly: The ambitious journalist discovered where we’d be going."
- **Scoring:**
Check for exact reproduction. Errors (e.g., phonemic substitutions, deletions, or distortions) indicate issues with morphosyntactic processing.
---
## Notes on Evaluation
TAB scoring is based on **binary pass/fail** criteria according to predefined aphasic markers. The evaluation focuses on the following linguistic features:
- **Anomia:** Word-finding failures or circumlocutions.
- **Paraphasias:**
- *Semantic Paraphasias:* Substitution of one content word for another.
- *Phonemic Paraphasias:* Errors in sound production (substitution, insertion, deletion, or transposition).
- **Agrammatism:** Omission of function words or morphemes.
- **Paragrammatism:** Incorrect or inappropriate grammatical structures.
- **Empty Speech:** Overuse of vague, nonspecific words.
- **Repetition/Reading Errors:** Deletions, insertions, or distortions.
For **Connected Text**, each response should be evaluated for the presence (1) or absence (0) of the following APROCSA features:
- Anomia
- Abandoned utterances
- Empty speech
- Semantic paraphasias
- Phonemic paraphasias (evaluated at token-level in models)
- Neologisms
- Jargon
- Perseverations
- Stereotypies and automatisms
- Short and simplified utterances
- Omission of bound morphemes
- Omission of function words
- Paragrammatism
- Retracing
- False starts
- Conduite d’approche
- Meaning unclear
- Off-topic
- Overall communication impairment
---
## Automatic Identification Prompt
The TAB is built for automatic evaluation using in-context LLM prompting. The system processes a transcript and produces a JSON file with each linguistic feature as a key and its value set to **1** (feature present) or **0** (feature not present).
### Instructions
- **Input:**
A transcript passage.
- **Processing:**
Analyze the transcript according to the definitions and examples below.
- **Output:**
A JSON file that includes **every one** of the following features as keys:
- Anomia
- Abandoned utterances
- Empty speech
- Semantic paraphasias
- Phonemic paraphasias
- Neologisms
- Jargon
- Perseverations
- Stereotypies and automatisms
- Short and simplified utterances
- Omission of bound morphemes
- Omission of function words
- Paragrammatism
- False starts
- Retracing
- Conduite d’approche
- Meaning unclear
- Off-topic
- Overall communication impairment
### Output Format
Your JSON output **must** include each feature as a key with its corresponding value (0 or 1). Do **not** include any additional keys or extraneous information.
---
### Examples
#### Example 1
**Hypothetical Passage:**
> "I was trying to tell you about my day but I just I mean I wanted to say something about the store I go store I wanted a pen I mean pencil The ball the ball the ball kept bouncing and I just stopped you know I keep saying dammit dammit dammit all the time"
**Associated JSON Output:**
```json
{
"Anomia": 1,
"Abandoned utterances": 1,
"Empty speech": 0,
"Semantic paraphasias": 0,
"Phonemic paraphasias": 0,
"Neologisms": 0,
"Jargon": 0,
"Perseverations": 1,
"Stereotypies and automatisms": 1,
"Short and simplified utterances": 1,
"Omission of bound morphemes": 1,
"Omission of function words": 1,
"Paragrammatism": 0,
"False starts": 0,
"Retracing": 0,
"Conduite d’approche": 0,
"Meaning unclear": 0,
"Off-topic": 0,
"Overall communication impairment": 1
}
```
#### Example 2
**Hypothetical Passage:**
> "I want to go to the store to buy a blorf You know I keep trying to say it but I say I want to go to the st store I want a pa pen I mean pencil I dont know what im trying to say It all seems not right"
**Associated JSON Output:**
```json
{
"Anomia": 1,
"Abandoned utterances": 0,
"Empty speech": 0,
"Semantic paraphasias": 0,
"Phonemic paraphasias": 0,
"Neologisms": 1,
"Jargon": 0,
"Perseverations": 0,
"Stereotypies and automatisms": 0,
"Short and simplified utterances": 0,
"Omission of bound morphemes": 0,
"Omission of function words": 0,
"Paragrammatism": 0,
"False starts": 1,
"Retracing": 0,
"Conduite d’approche": 1,
"Meaning unclear": 1,
"Off-topic": 0,
"Overall communication impairment": 1
}
```
> **Important:** Your analysis must strictly adhere to the definitions and examples provided. No additional keys or information should be included in the output JSON.
---
## Sources
1. **Wilson SM, Eriksson DK, Schneck SM, Lucanie JM.**
*A quick aphasia battery for efficient, reliable, and multidimensional assessment of language function.*
**PLoS One** 2018; 13(2): e0192773.
2. **Casilio M, Rising K, Beeson PM, Bunton K, Wilson SM.**
*Auditory-perceptual rating of connected speech in aphasia.*
**Am J Speech Lang Pathol** 2019; 28: 550-68.
*Definitions and examples for the automatic identification prompt are adapted from Table 1 of Casilio et al. (2019).*
---
## Disclaimer
The TAB is intended **for research and evaluation purposes only**. It should not be used as a clinical diagnostic tool or a substitute for professional evaluation.
---
## Contributing
Contributions are welcome! If you wish to improve or extend TAB:
1. Fork the repository.
2. Create your feature branch.
3. Commit your changes.
4. Open a pull request.
---
## License
the TAB is released openly under the MIT license