Demonstrating Par4Sem - A Semantic Writing Aid with Adaptive Paraphrasing

Abstract

In this paper, we present PAR4SEM, a semantic writing aid tool based on adaptive paraphrasing. Unlike many annotation tools that are primarily used to collect training examples, PAR4SEM is integrated into a real word application, in this case a writing aid tool, in order to collect training examples from usage data. PAR4SEM is a tool, which supports an adaptive, iterative, and interactive process where the underlying machine learning models are updated for each iteration using new training examples from usage data. After motivating the use of ever-learning tools in NLP applications, we evaluate PAR4SEM by adopting it to a text simplification task through mere usage.

A Design Space for Intelligent and Interactive Writing Assistants

“…Rewriting problems involve sequence transduction tasks, where texts from one form are transformed to another while improving the quality by making them fluent, clear, readable, and coherent. These tasks are essential in various writing assistance applications, such as grammatical error correction [37,43,277], paraphrasing [267], or general-purpose text editing [66,70,204,226] to name a few. Generation refers primarily to problems that involve the creation of new, contextually relevant, coherent, and readable text from relatively limited inputs, such as autocomplete, paraphrasing, and story generation [7,45,117,220].…”

“…Recently, models have been developed with access to additional tools or data at inference time to make them capable of providing assistance beyond the knowledge encoded in their parameters. In the case of tool, a model may access external software or third-party APIs to perform tasks like search, translation, or calculator, or even setting calendar events on behalf of users [43,180,267,272]. Data refers to external datasets or resources, such as information stored in a database, external knowledge repositories, or any other structured/unstructured data sources that the models might leverage to provide writing assistance [133,223,230,276].…”

“…as colors and underlines [58,267,268]. Another common mechanism is to keep system outputs in a separate location like a wizard or panel [57,195,252].…”

“…With developments in model architecture such as statistical ML models and deep neural networks [51,172], models behind writing tools could take advantage of larger sources of unlabeled data. For instance, some work started using learned word embeddings or trained embeddings specifically for a task [56,84,267]. They were then able to bootstrap off this to use smaller human-labeled datasets for further modeling and evaluation.…”