Go Back
Deep Learning

Generation

Natural Language Generation (NLG) is the process of using algorithms to generate human language. This can include everything from sentences and paragraphs to nuanced words and tenses. However applied, the goal remains the same: turn ones and zeros into a natural-sounding, coherent language narrative that’s easily read and received by the intended audience(s).

NLG, and even more specifically Natural Language Processing (NLP), is often seen as a subset of Machine Learning. The concept and algorithms have existed for decades, but only recently has NLG thrived in commercial applications, such as chatbots, journalism, summarized medical records, and auto-generated product descriptions – to name a few. Prior to this renaissance, most hindrances were the result of sparse data sets and a lack of appropriate use cases.

Data is primarily sourced from public corpora (such as subpoenaed email records), web scraping techniques (i.e. combing text from millions of websites), and private data sources (such as internal records). Without an established use case, wrangling this data is often challenging. In modern times, when NLG is applied to a single industry, domain, or vertical and has a specific end-user outcome, it has outpaced NLG applications that apply to broad audiences and take a horizontal approach. Encouragingly, recent advancements suggest a convergence may be coming.

Use Cases

Capturing the art of fundraising language is at the core of what Gravyty does best.

For example, we look at past interactions (emails, contact reports, and more) to create customized language models that pair fundraising best practices with contextually relevant data and timely insights. As a result, we produce personalized communications that are indistinguishable from those of seasoned gift officers.

Time is money. Researching and writing effective email copy by hand takes a lot of time. We’ve found that fundraisers often spend half their day pulling together data – looking back at past communications, thinking about what to write, fine-tuning messaging, proofreading, and fact-checking – to produce a few lines of an email. This is especially true for new fundraisers still in the onboarding phase of their positions. Gravyty’s First Draft platform uses NLG to automate this entire process for the fundraiser by delivering email drafts directly to their inbox, enabling users to build relationships at scale, thus freeing them up to spend more time focusing on inspiring gifts.

Resources

Word2VecCapture semantic properties of language by mapping words and phrases to vectors of real numbers

LINK

A Primer on Neural Network Models for Natural Language Processing

LINK

Exploring the Limits of Language Modeling

LINK