Natural Language Processing Q&A
In this Q&A, Jos Martin, Senior Engineering Manager at MathWorks, discusses recent NLP developments and the applications that are benefitting from the technology.
Sponsored Post.
The Changing Nature of Natural Language Processing: How Will It Be Applied in the Future?
Natural language processing (NLP) software has been used for years to enable speech recognition in everything from cars to smartphones, and more recently, smart home devices. New advances in NLP continue to refine the technology for an expanding array of applications such as sentiment and recommendation analysis, predictive text modelling and the emerging field of transfer learning.
In this Q&A, Jos Martin, Senior Engineering Manager at MathWorks, discusses recent NLP developments and the applications that are benefitting from the technology.
KDnuggets: How has NLP changed over the last five to 10 years?
Jos Martin: The biggest change surrounding natural language processing has been the shift away from traditional text and word pattern recognition toward approaches that incorporate machine learning and deep learning. This allows semantic meaning to be transferred into the numerical space, which makes many recently developed machine learning and deep learning techniques available to NLP system designers.
For example, many systems take advantage of recurrent neural networks that rely on long short-term memory (LSTM). Rather than simply processing single data points, or words, LSTM can help us learn more about the relationships between words in sentences, paragraphs and other linguistic blocks. These types of networks allow designers to predict what the next word in a sequence might be or determine probabilities for the next several words.
Another area where there is significant change is the application of NLP to speech. Pairing speech recognition and NLP models to guess which words go together ultimately creates better speech recognition models overall.
KDnuggets: Can you describe some current applications that are using NLP?
JM: One application is sentiment analysis, which analyzes a block of text to understand how much it expresses a specific set of sentiments (i.e. numeric agreement or disagreement with pre-defined classifications trained into the model). This is also broadening to encompass more than just what might be termed “human sentiments.”
A second use case lies in predictive text models, which can be joined with a speech recognition model to significantly improve the behavior of the overall model. This is evident in how NLP is being used by chatbots and virtual assistants to help users more quickly complete specific tasks. For example, Google Assistant can autocorrect speech-to-text errors such as “Set my alarm for three thirds” to “Set my alarm for three-thirty.”
Recommendation analysis is another application that is adopting NLP, in this case to enable the analysis of unstructured text, such as limited information about a product or service. Here, NLP model designers train deep networks to process unstructured text and infer something about the product, particularly what other products it resembles, and which other products customers might be interested in exploring.
Machine translation – also known as automatic language translation – has also improved over the last few years given advances in NLP and deep learning models.
KDnuggets: Where should we expect NLP to be applied in the future as it continues to evolve?
Enhanced speech recognition will be a big area for NLP. While speech recognition has become significantly more accurate over the last several years, there are still ways the technology must improve before it can be used more broadly. Some of the advances in this area will come from better NLP models.
Another area will be the application of transfer learning to NLP, where a model developed for one task can be retrained for another problem. Only recently has it become possible to apply such pre-trained models to NLP. This is important because it allows users to take one pre-trained model, modify it slightly, and apply it across a variety of other text analytics tasks, such as sentiment analysis or question answering, where designers try to automatically answer questions that people ask using natural language.
NLP can also potentially be applied to document summarization. In the future, NLP models will not only summarize the sentiment of a collection of text, they will be able to summarize the content, too.
The above Q&A provides some insight into the future of NLP. To learn more about NLP and how MathWorks is meeting this new trend see the links below or email me at jlmartin@mathworks.com.
- Natural Language Processing: This web page provides an overview of NLP along with examples.
- Extract Text Data from Files: This example shows how to exact the text data from text, HTML, Microsoft Word, PDF, CSV, and Microsoft Excel files and import it into MATLAB for analysis.
- Create Simple Text Model for Classification: This example shows how to train a simple text classifier on word frequency counts using a bag-of-words model.
- Classify Text Data Using Deep Learning: This example shows how to classify text descriptions of weather reports using a deep learning short-term memory (LSTM) network.