Natural language processing examples – Virtual assistants
Valérie Bécaert Valérie Bécaert
March 3 6 min

Natural language processing examples – Virtual assistants

Natural language processing (or NLP) is perhaps one of the biggest success stories in AI technology. As little as a decade ago, most people would have viewed the idea of machines understanding language as sci-fi-esque and futuristic. Today, the technology is widespread, and most of us use it in our daily life, often without knowing it. That’s because with the rise of machine learning and artificial intelligence, the challenges associated with processing natural language can now be managed much easier than before. And in terms of natural language processing examples, there’s no better place to look than the rise of virtual assistants like Cortana, Siri, and Alexa.

In this blog, we discuss the challenges of natural language processing and how virtual assistants have overcome them.

What is natural language processing?

Natural language processing refers to algorithms that allow computers programs to understand human language. Natural here refers to an organically evolving language, like Spanish, rather than a constructed language like Klingon, or a computing language like JavaScript. The key here is the word understand.

Machines have interacted with language based on rudimentary guidelines for decades, without understanding the semantic intent of the text. Think of traditional keyword searchers, or spellcheckers. They might be able to match words that look the same or recognize that a word ending in “-ing” is a verb — but the understanding doesn’t go any deeper than that.

Natural language processing found in a virtual assistant needs to understand the words you’re using to process them. That presents a raft of new challenges that, until recently, algorithms weren’t sophisticated enough to negotiate.

The challenge of natural language processing

Natural language processing poses very similar challenges to machine translation. It makes sense therefore, that the two technologies have improved in line with one another over the past few years.

The reason that it’s difficult for computers to understand us is because language is complicated, and there’s a lot of it. Think of the sheer amount of words in a single language, then multiply that by all the languages in the world. Then, consider the different ways that meaning is conveyed outside of words – grammar, syntax, semantic, register, dialect, irony … the list goes on. It’d be impossible for humans even to quantify rules to govern all of this, never mind teach it to an algorithm.

Because the challenges associated with natural language processing are so numerous, differing, and complex – so too must be the solutions. There’s no single algorithm that can interpret language. However, a more complex combination of different algorithms with differing objectives working in harmony can provide the answer.

Natural language processing examples: How do virtual assistants work

Though the voice search found in virtual assistants is the most well-known deployment of NLP, there are plenty of other deployments available. But whatever NLP is being used for, the first step is to convert the speech into text that the algorithm can understand before it can be processed. This is done with a speech-to-text algorithm.

It might seem like converting voice to text is the simplest part of the process. But in fact, it’s something that machines have struggled with for years – and was one of the most significant obstacles to effective natural language processing. In fact, it was deep learning that provided the key here, a type of machine learning. Now when you say, “Where’s the nearest restaurant?” into your virtual assistant, the machine has a reliable function that converts the audio information into something it can process.

Next comes the process of understanding the language. This comes in two broad stages: first analyzing the structure, and then the meaning.

Structure

In this stage, the virtual assistant analyzes the words in your sentence and the relationship between them to understand where the important information is. This means identifying where the verbs and nouns are, as well as other information. Using techniques like stemming and lemmatization, the computer boils your vocabulary down to its root meaning, turning phrases like “was going”, “nearest”, and “cats” into “go,” “near”, and “cat.” In linguistic terms, this means removing inflections, conjugations, and declensions. This simplification of language makes it easier to analyze later in the process.

After this, techniques like word segmentation and sentence breaking are used to split the text into analyzable chunks, and parsing will undertake a grammatical analysis of the resulting sections.

Semantics

Semantics is a linguistic term referring to the meaning of a word. It’s what allows people to associate the sounds and letters that make up the word “dog” with the furry four-legged creature we all know.

In this process, techniques such as named entity recognition are used to determine which parts of the text can be identified and categorized into preset groups. Again, if you ask where your nearest restaurant is, that involves inferring “distance” and “restaurant/food” from your speech. This boils down a relatively complex sentence into actionable categories that search engines have been able to process for several years.

But obviously not everything is that simple. Meanings often change depending on register, meaning, and context. For that, the system must employ word sense disambiguation and natural language generation to infer more complex semantics, based on the context and wider structure of a sentence. These processes allow voice assistants to convert commands like “Where’s good to eat around here” into the same “distance” and “restaurant” elements as a simpler command like ”Find my nearest restaurant.”

Once these processes have been undertaken, it’s easy for the voice assistant to interact with applications or search engines in the same way they always have. Once the result has been reached the algorithms are used in reverse to convert that data back into understandable human language.

A future for natural language processing

Perhaps the most fascinating thing about natural language processing is the way it’s changed the fundamental way we interact with technology, through typing information, into something more seamless and dynamic. As the technology improves, we’ll see even more profound changes to the ways it's used and see our traditional relationship with technology transforming yet further.

NLP has made significant strides forward in the last few years – but there’s plenty of distance still to go. The question for now is: where do we travel next?