3 NLP Trends on the Rise in 2018

With advances in computational power and the integration of artificial intelligence, the natural language processing domain has evolved into a whirlwind of innovation. In fact, experts expect the NLP market to swell to an impressive $22.3 billion by 2025. In the meantime, here’s a look at three NLP trends that are poised to accelerate over the course of this year.

Chatbots

According to the International Data Corporation, searching for information, whether it be from a database or the internet, steals a substantial 30% of the typical “knowledge worker’s” day. With that theft comes a loss of productivity as workers dedicate time to tasks that are menial relative to the big picture. Office assistants, particularly in the form of chatbots, however, offer a chance to disrupt data search and retrieval as we know it. Chatbots can streamline project management by making it easier to access, organize, and analyze data. They also allow for routine data-intensive tasks to be automated. With the annual growth rate of chatbot revenue estimated between 24.3% and 37%, we can anticipate more AI-fueled bots making their way into the market going forward.

Corporations and start-ups alike strive to make their mark in this rapidly expanding vertical. Some of the most notable administrative chatbots in the enterprise space at the moment include: Microsoft’s Cortana, Amazon’s Alexa for Business, IBM’s Watson assistant, Google Assistant, and Apple’s Siri.

More recently on the newcomer front, a new AI-powered office assistant by the name of Jane launched to $8.3 million worth of fanfare — funding, that is. What sets Jane apart from its competitors is its reliance upon a human “copilot” for those confounding questions it can’t seem to answer. Even better, Jane proceeds to enrich its own knowledge base with these employee-supplemented responses so it need never call for backup again. As Jane’s abilities demonstrate, today’s bots are getting smarter with every query.  

Beyond administrative assistance, chatbots demonstrate utility in the customer service realm. Frequently asked questions and simple lookup tasks can be relegated to automated helpers. This leaves customer service agents free to devote time to troubleshooting bigger matters that personalize and enhance the customer experience. On the customer-facing end as well as internally, chatbots can save valuable time and energy for all members of the value stream. Chatbot technology is poised for considerable growth as speech and language processing tools become more robust by expanding beyond rules-based engines to include neural conversational models.

Neural machine translation

While statistical phrase-based translation has enjoyed a stretch of popularity, the technique’s focus on sentential constituents means that it struggles when translating between languages with considerably different word orderings. How to resolve this issue? Enter neural machine translation (NMT).

Using a neural network model, NMT’s advantage is that “a single system can be trained directly on source and target text, no longer requiring the pipeline of specialized systems used in statistical machine learning.” In addition to this, NMT diminishes the required amount of post-translation editing by 25% and generally enhances the fluency of translations. Unlike statistical machine translation, NMT can capture word similarity, local context, and relationships between specific languages. Although NMT necessitates a greater amount of training data, major companies such as Google and Facebook are jumping onboard in pursuit of more accurate translations that are sensitive to the subtleties of language.

Google now offers offline NMT for 59 languages in its Translate app, a welcome step away from the older phrase-based translation system that was previously in place. According to Google, “The neural system translates whole sentences at a time, rather than piece by piece. It uses broader context to help determine the most relevant translation, which it then rearranges and adjusts to sound more like a real person speaking with proper grammar. This makes translated paragraphs and articles a lot smoother and easier to read.”

Last year, Facebook made a complete shift to using NMT as the backbone of its automatic translation system for online content. As a result, Facebook’s new translation framework increased its BLEU (bilingual evaluation understudy) score — a metric that is the standard for machine translation assessment — by approximately 11% across all languages. In the midst of all this, a recent spike in research papers on NMT provides further evidence that the NLP world is currently riding a wave that shows no signs of crashing.

News verification

You can hardly read the news these days without hearing about the encroachment of falsehoods and fake news upon the media. Suddenly the veritability of even the most trusted sources is being called into question. As the Internet churns out content at breakneck speed and disseminates it far and wide, research has turned to AI as a potentially scalable solution to counter the proliferation of false information. If we can train models to grasp what fake news looks like, perhaps we can weed out sites and companies that are its largest propagators — whether intentionally or through lackadaisical fact-checking.

Factmata, based in the UK, is one of a number of companies that is harnessing machine learning to combat the spread of disinformation. It goes deeper than metadata to assess content, looking closely at the language and logic being wielded. Though Factmata has yet to debut a product, the company is testing out three main services: (1) detecting illegitimate ads, (2) flagging biased stories on news aggregation platforms, and (3) providing consumers with information about news pieces in real-time. Ideally, Factmata’s strategy will consist of a team of expert users who perform fact-checking, mobilized in tandem with AI.

Factmata founder and CEO Dhruv Ghulati received a CogX 2018 Rising Star award from the UK prime minister for the company’s work toward advancing truth in media. Ghulati noted that the biggest players in fake news are “often financially driven—fake news stories correlate highly from a linguistic perspective to clickbait stories.” Using computational methods to recognize the language typical of these non-credible stories could provide an avenue for defunding the sites that foster them.

After facing backlash for the role its site has played in spreading fake news (particularly in the 2016 U.S. presidential election), Facebook released an announcement that details its elaborate fact-checking efforts. The company noted the utility of machine learning in identifying “duplicates of debunked stories”  and “foreign Pages that are likely to spread financially-motivated hoaxes to people in other countries.” While Facebook and some other automated fact-checkers still rely partially upon human review, advances in computer science are increasingly turning verification into a popular NLP task.

News verification, chatbots, and neural machine translation may seem to have little in common, but all of these up and coming developments exemplify an ongoing shift in NLP towards machine learning and AI. Researchers in the field have made significant progress in the past few years, but the intricacy of language means there remain many more mountains to climb. These 2018 trends are merely a snapshot of what is to come.