Natural Language Processing (NLP) is transforming the way humans interact with technology, enabling machines to understand, interpret and respond to human language. As advancements in machine learning and artificial intelligence continue to accelerate, NLP is expanding across industries, from voice assistants and chatbots to healthcare, automobiles and entertainment. This article explores the NLP patent landscape, highlighting key technologies, major applications, industry adoption and emerging market opportunities. By examining patent trends from 2017 to 2019, it reveals how innovation is shaping the future of NLP and where the next breakthroughs may emerge.

What Goes In: Text, Speech, Images and Gestures

Text is the dominant input, with 968 patents using it, followed by 473 that use speech. Looking at patents that rely on only one type of input gives a cleaner picture:

Input type	Patents using only this input
Text only	635
Speech only	147
Image/video only	20
Gesture only	2
All input types together	18

 

The 18 patents that accept every input type are an interesting group. The earliest has a priority date in 2002, showing that the idea of fully multimodal interaction is not new. Interest matured between 2014 and 2017. Four of these 18, filed by Apple, Microsoft, Pearson LLC and Verint Systems, are designed to work across all three server types (local, external and cloud).

Gesture-only input, with just two patents, is the most open space of all. As augmented reality and wearable devices spread, gesture and multimodal language interfaces may become more relevant.

How It Works: The Rise of Machine Learning

Summing the technology-by-purpose map across 2017 to 2019 shows how heavily the field leans toward learned models.

Purpose	Rule-based	Machine learning	ML share
Text analytics	212	883	81%
Assistance	45	259	85%
Sentiment analysis	17	136	89%
Speech recognition and analytics	19	130	87%
Web browsing	13	70	84%
Others	9	64	88%

Totals are sums of the yearly counts in the study’s technology map; ML share is calculated here. A patent can appear in more than one category.

 

Machine learning leads in every category. IBM is the largest filer of machine-learning-based patents with 167 and the US leads with 515 families against China’s 204. Rule-based methods have not disappeared: 102 patents are hybrids, reflecting the practical need for controllable behaviour.

The yearly pattern is instructive. Machine-learning text analytics rose from 240 in 2017 to 471 in 2018. The figure of 172 for 2019 looks like a drop, but the data window ends in June 2019 and many recent filings are not yet published, so it should not be read as a decline.

A note on what happened next

The study’s window closed just as a major shift was under way. The Transformer architecture was introduced in 2017 and models such as BERT and GPT-2 followed in 2018 and 2019. Since then, large language models have changed what NLP systems can do, from fluent text generation to reasoning over documents. Patent trends after 2019 would very likely show a sharp rise in filings around generative models, retrieval and conversational agents. The findings here are best read as the foundation on which that wave was built.

What It Is For: Applications

Text analytics

Text analytics is the largest area. Within it, data identification and modelling leads (443 applications, 149 granted), followed by data classification (313 and 114), data filtering (53 and 23), question answering (131 and 47) and machine translation (43 and 13).

Speech analytics

Question answering is the largest speech application at 65 published applications (12 granted), ahead of machine translation (22 and 7) and language identification (16 and 10). Machine-learning question answering over speech grew from 14 in 2017 to 26 in 2019, even as the part-year data caveat applies.

Assistance

Application	Applications	Granted	Approx. grant share
Mobile assistant	80	43	54%
Customer support	57	28	49%
Chatbot	52	7	13%
Automobile assistant	15	2	13%
Other	33	9	27%

Grant share is calculated here as granted divided by applications from the study’s charts. Because categories can overlap and the two counts are not strictly the same cohort, treat it as indicative.

 

Mobile assistant patents are the most mature, with the highest proportion granted. They were also concentrated in 2017 (78 ML filings) and have tapered since, which is typical when the first wave of a technology matures. Customer support patents total 85 and are mostly filed in the US and China, with IBM first and Accenture second. By contrast, chatbots and automobile assistants have many pending applications but few grants, which suggests newer and less settled territory.

Sentiment analysis and web browsing

Sentiment and emotion analysis has 128 applications and 45 granted. US filings lead with 99, followed by China with 45. Web browsing has 45 applications and 29 granted, a high grant share but a smaller and slower-growing area.

Where the servers are

Just 28 patents can operate across cloud, local and external servers. About 82% of them (roughly 23) cover US jurisdiction, with most filed in 2016. Beyond the US, Australia and South Korea are the other jurisdictions exploring all server types. As privacy and latency concerns push more processing onto devices, hybrid deployments across cloud and local servers are worth watching.

When Other Industries Join In

One of the most useful insights from the study is that NLP is escaping the pure-technology sector. Four examples illustrate the trend.

• Automotive: Honda’s US9747898B2 interprets ambiguous vehicle instructions from occupants. Toyota filed three patents (JP2018173846A, US9997160B2, CN106919997A) integrating machine translation, language identification and smart commands into cars. Carmakers are building their own portfolios instead of licensing from others.
• Healthcare: Dr Systems Inc. filed US2019147346A1 for a conversational interface that uses deep learning and medical knowledge to track the state of a conversation and surface relevant information.
• Media and entertainment: Xian Film Television Data Evaluation Center filed CN108563688A and CN108549630A for sentiment analysis of film and television scripts.
• Energy and aerospace: State Grid’s 2019 collaborative patent and the Avic–Xidian aircraft script control patent show NLP spreading into industrial control.

Market Outlook and Opportunities

At the time of the study, the NLP market was expected to reach USD 16.07 billion by 2021. Later industry forecasts have generally been much higher, particularly after the arrival of large language models, but exact figures vary widely by source and definition. The direction of travel is consistent across all of them: strong growth.

The study points to several demand drivers:

• Voice assistants and chatbots: with every major technology company launching a personal assistant, the race to make them more human-like will continue.
• Automobile assistants: still in early stages, but accelerating with electric and connected vehicles as manufacturers compete to define the in-car experience.
• Healthcare NLP: growth is expected to be exponential, driven by the need to manage clinical data efficiently, such as extracting information from notes and records.
• Entertainment and media: early patents show script and audience analysis as a new area.

Where the white space is

• Gesture and multimodal input, with only 2 gesture-only patents and 18 using all input types.
• Automobile assistance, which has few grants and a small base of filings.
• Mobile assistance in Japan and Canada, where the study found little activity.
• Under-served jurisdictions, including Sweden, the United Kingdom, Taiwan and Germany.
• Collaboration between companies and universities, especially outside China.

Cautions

A landscape study can show where filings cluster, but it cannot tell you whether a specific product infringes or is free to operate. One patent in the data set, US10057733B2, assigned to an individual inventor, is under litigation, a reminder that active enforcement exists even in a young field. Anyone acting on these insights should commission a focused freedom-to-operate analysis.

Conclusion

Across this series, a consistent story emerges. NLP patenting between 2017 and 2019 was concentrated in the US and China, led by large technology companies, increasingly built on machine learning and centred on text. The most mature applications are mobile and customer-support assistants, while automobiles, healthcare, entertainment and multimodal interfaces look like the next opportunities. For innovators and investors, the main lesson is to look beyond sheer patent counts: influence, collaboration and unexplored niches matter at least as much.

Source: Landscape Study – Natural Language Processing (patent data 01/06/2017 to 01/06/2019). Percentages and context on later developments are added by the author of this article.