This project is currently under development by Tev'n Powers to satisfy the Master's Thesis component of the University of Washington's Master's in Computational Linguistics. Special thanks to my thesis committee: Cecilia Aragon (advisor) and Emily Bender (reader). To share any questions, thoughts, or concerns please contact tevn@uw.edu or make a GitHub issue.

TextStudio

TextStudio is a text processing architecture comprised of the text-studio Python package, software development kit (SDK), and desktop application. Each of these components contributes to creating the development environment where users can explore, process, model, and visualize textual data.

TextStudio is designed for the following audiences and purposes:

1. Experienced natural language processing practitioners (e.g. computational linguists, software engineers, data scientists, etc.) benefit from TextStudio's improvement to the current developer experience by removing inefficiencies in the experimentation workflow. Built on the premise of collaboration, TextStudio simplifies discovery of NLP, statistical modeling, and machine learning modules for text data and enables seamless module integration into text processing pipelines via a plugin extensible architecture.
2. For users with minimal software engineering or computational linguistics backgrounds, TextStudio provides access to a rich set of text processing systems and techniques developed by the NLP community. Using the visual editor in the desktop application, every user has the power build sophisticated text processing pipelines to extract insights from any text data set without writing a single line of code.

Architecture

Desktop Application (coming soon)

The TextStudio application provides a visual interface for users to inspect and visualize text data, as well as facilitates the building of text processing systems based on NLP techniques and statistical or machine learning models.

Software Development Kit (SDK) (coming soon)

The software development kit is built upon the text-studio Python package and enables users to write and publish plugins that can be used by users of the desktop application.

Installation (coming soon)

pip install text-studio

From source: git clone https://github.com/tevnpowers/text-studio

Plugin Components

The major value of TextStudio is the hackable design that enables the developer community to extend its core functionality to support:

• Loading data from arbitrary sources
• Providing annotations on text data instances
• Generating artifacts (summary reports, statistics, visualizations, etc.) from text
• Modeling text data

Data Loader

In this system, a Data Loader is responsible for loading data that exists outside of the application into a canonical TextStudio data set. It must also provide the inverse functionality, to write a TextStudio data set to an external location.

A DataLoader plugin is a subclass of the text_studio.DataLoader abstract class, and implements the following class methods:

• load: Create the list of data instances for a text_studio.Datasetfrom data that exists outside of the application.
  • Parameters:
    • file_path: The path to a data file or directory which contains data files to be loaded into a data set.
    • **kwargs: Additional keyword arguments that the author can optionally require in order to configure the load logic.
  • Return:
    • List of dictionary objects, each of which represents a single data instance in the data set.
• save: Export the list of data instances in a text_studio.Dataset to a storage system outside of the application.
  • Parameters:
    • file_path: The path to a data file or directory where the data set should be exported.
    • **kwargs: Additional keyword arguments that the author can optionally require in order to configure the save logic.
  • Return:
    • Boolean value that is True if a data set is successfully exported and False if the save failed for any reason.

Pipeline

A text processing pipeline is any combination of Annotator, Action, or Model components that run in a sequence on an input data set. Pipelines themselves are defined in text_studio.Pipeline and generated by the TextStudio desktop application, not by developers.

However, developers may write plugins for each pipeline component type further described below.

Annotator

An Annotator runs a process which augments the input data it is given. That is, given a data instance object (Python dictionary), an annotator will add a new key value pair to the dictionary (e.g. tokenization output, part of speech tags, lemmatized version of the raw text, etc.).

An Annotator plugin is a subclass of the text_studio.Annotator abstract class, and implements the following class methods:

• __init__: Configure the settings needed for the Annotator module to properly function.
  • Parameters:
    • keys: the list of keys (strings) in the data instance object dictionary correspond to the values that the Annotator needs in order to extract the data required for execution.
    • annotations: the list of keys (strings) that an Annotator should add to the data instance object dictionary, where the corresponding value(s) are computed by the Annotator when executed.
    • Additional Named Arguments: A plugin author may require any arbitrary named arguments that are necessary to configure the module's execution.
• process_single: Annotate a single data instance with a new value.
  • Parameters:
    • doc: A dictionary representing a single data instance.
  • Return:
    • A dictionary that is the augmented version of the input object, now annotated with additional information.
• process_batch: Annotate a collection of data instances with new values.
  • Parameters:
    • docs: An iterable containing dictionaries, which each represent a single data instance.
  • Return:
    • A collection of dictionaries, where each is an augmented version of an input object, now annotated with additional information.

Action

An Action consumes input data either individually or in bulk in order to produce an artifact about the input data, while not modifying or augmenting the input data instance(s). In this case, an artifact may be a visualization, a summary report, or any other insights that can be extracted from the provided data.

An Action plugin is a subclass of the text_studio.Action abstract class, and implements the following class methods:

• __init__: Configure the settings needed for the Action module to properly function.
  • Parameters:
    • keys: the list of keys (strings) in the data instance object dictionary correspond to the values that the Action needs in order to extract the data required for execution.
    • Additional Named Arguments: A plugin author may require any arbitrary named arguments that are necessary to configure the module's execution.
• process_single: Process a single data instance to produce insights.
  • Parameters:
    • doc: A dictionary representing a single data instance. information.
• process_batch: Process a collection of data instances to produce insights.
  • Parameters:
    • docs: An iterable containing dictionaries, which each represent a single data instance.

Model

(coming soon)