Natural Language Understanding (NLU) applications have been at the center of our world for years, whether we realize it or not. Digital assistants such as Alexa and Siri have used NLU to divine human intent with varying levels of success for well over a decade. Despite some frustrating starts, NLU is seeing a renaissance with many Generative AI applications that not only complete a human’s sentence but also write entire articles or even compose art. While we may not be at the point where AI chatbots are self-aware, they are clearly getting closer to passing the AI intelligence gold standard that is the Turing Test.
There are currently many generative AI services available with the number growing every day. Many of these services are becoming more accurate, useful, and readily available for the average consumer. For example, you may have seen an application that generates realistic pictures of people who don’t exist, allowing prospective internet trolls a better way to disguise their identity online. An application named craiyon leverages technology introduced in OpenAI’s Dall-e project to create on-demand art in any context or style. Perhaps the most widely known is OpenAI’s Generative Pre-trained Transformer 3 (GPT-3) project. GPT-3 uses deep learning to produce convincing, human-like text that is, in many cases, indistinguishable from the real thing.
These advances are remarkable steps forward, but what if AI predictive capabilities could be applied to financial technology applications? What if you could just ask for what you wanted in a FinTech app instead of learning a series of filters or a potentially complex formula language? In the world of financial services, there is much to be gained by leveraging inefficiencies in markets through technological time savings. NLU is an extremely promising technology that can save time for financial professionals who typically spend countless hours poring over financial data and documents to make informed decisions.
Many financial applications use a language and lexicon all their own. Complex financial models and a nearly inexhaustible amount of data require that FinTech companies create their own proprietary programming languages to allow their clients to get the most customization and flexibility out of this data, e.g., FactSet’s Query Language (FQL). Some of these languages have been around for decades and have become quite complex. While these languages are compelling, they are not always friendly to newer users, and the skills learned may not be transferrable from one language to the next. In the future, NLU will play a huge role in overcoming these challenges so that financial professionals will merely have to ask for what they need instead of learning how to code it.
Despite the amazing advances in the products mentioned at the beginning of this article, there are still many obstacles to asking for what you want and receiving a machine-produced result. Foreign language translation is one of the largest stumbling blocks, particularly for queries that do not begin in English. Formula languages all begin with a standard language and syntax. However, when you ask for what you want in your language, the phrase must be parsed and translated before NLU can determine the intent. This process adds many potential points of failure, and countless regional nuances can get lost in translation.
However, there is hope. While we are nowhere near a Star Trek-style universal translator yet, translation services are getting better and more accurate every day. However, translating from French to English and then back to French still has inherent problems and could cause a giant game of telephone where a machine plays many of the roles in this game.
Natural Language Understanding and its output equivalent, Natural Language Generation (NLG), have been paired together in the last few years to turn human text into software code. Developers can now use natural language to create video games on the fly or conduct more traditional pair programming exercises.
In financial applications, NLU and NLG offer non-engineers the time-saving shortcut of writing the code. While the code still needs to be executed to reach a tangible result, NLG provides an output where the act of writing code is no longer a middleman that can hold up the process. This makes the financial formula language learning curve much shallower. For example, if you were to query:
"Show me companies with sales greater than the average sales of companies in the S&P 500.”
An NLU/NLG service could respond with not only a list of companies that meet the specified criteria but also the code used to derive that list:
FF_SALES(ANN,0)>UAVG(FG_CONSTITUENTS(SP50,0,CLOSE),FF_SALES(ANN,0))
For many people, this line of code will explain the formula language elements and syntax better than any online help documentation. This output teaches how the result was created and can reveal ideas and shortcuts that were previously unknown. The code can then be tweaked and re-arranged to give the exact result needed. This discovery and reorganization process can ultimately help you arrive at a more complex and valuable result that neither the query language nor the NLG could easily deliver alone.
FactSet’s Cognitive Computing group continually explores NLU/NLG services to implement in applications. One recent example is in the Screening application where clients can now search for mutual funds and ETFs more easily using NLU. These Machine Learning services allow clients the ability to narrow down an extremely long list of funds into something more manageable and customized to their needs.
Source: FactSet
Our smartphone world has been a game changer in so many ways. As a result, many FinTech companies have taken the first step toward creating their own smart speaker-style interfaces. You may have already seen these interfaces in some FinTech mobile apps with the addition of a microphone in search fields. Bringing the power of NLU with you is the next step in the evolution of these services. While many people might think this functionality is commonplace these days, it is anything but. Voice translation brings an entirely new set of challenges to NLU that text does not have. Microphone issues, background noise, and bad translations are but a few issues that Machine Learning engineers need to address before any product in this space can be taken seriously and eventually trusted by financial professionals.
We have asked a lot of “What if?” questions in this article, but the truth is that these applications are here now, and they are only getting better. NLU and NLG are rapidly evolving areas of Machine Learning where we are just beginning to scratch the surface of what is possible within financial applications. The ability to write code wherever and whenever you want is no longer limited to the realm of professional software developers. This functionality is coming to a financial services application near you, and it will save you enormous amounts of time and effort.
This blog post is for informational purposes only. The information contained in this blog post is not legal, tax, or investment advice. FactSet does not endorse or recommend any investments and assumes no liability for any consequence relating directly or indirectly to any action or inaction taken based on the information contained in this article.