Some Thoughts About Voice Search Part I

Voice search is one of the most exciting fields in artificial intelligence(AI). Even before the raise in popularity of AI, the emergence of smartphones and wearable’s forced the industry to start envisioning new search experiences for those digital mediums. Now, the advancements in areas such as natural language understanding(NL) have open the door to new search experiences using voice as the fundamental user interface.

To understanding voice search, we should first agree to avoid drawing too many parallels with traditional web search techniques. Voice search experiences are fundamentally different that its web predecessors and are closer to the communication patterns we use to conduct inquires in normal conversations. To illustrate that point, I’ve put together a list of some of thekeyy unique characteristics of coice search experiences as well as its differences with traditional search mechanisms.

1 — Questions vs. Keywords

An obvious difference between voice search and web-search experiences is that in the former users will express the search using natural language questions compared to the traditional keyword mechanism of web search engines. This aspects is very relevant as there are infinite ways on which we can formulate the same questions in natural language which makes the understanding of the semantics of a questions a key element of voice search experiences.

2 — Stateful Searches

Imagine a scenario on which a users asks a digital assistant (DA) a question like the following:

User: Trinity( my chosen name for our DA ;) ), could you find all documents that refer to AA 2016 earnings?

Followed by User: And please include documents that include Wall Street predictions for the airline industry.

That type of interaction composes two searches into a single experience. A voice search engine should be able to maintain and understand the state of the conversation between searches.

3 — Voice and Data Responses

Voice search experiences should be able to produce responses via voice as well as data. In many cases, a voice search engine should generate short narratives that describe the results of a specific search.

4 — Clarifications

Let’s g back to our favorite DA.

User: Trinity, could you find the document supporting my Tax returns?

DA: Just for 2016?

User: Yes please

That previous dialog leverages a clarification question as a way to enrich the search criteria. Although the pattern is completely natural in human conversations is not very common in web search experiences.

5 — Interpretation

User: Trinity, what’s causing the market downturn today?

That type of questions requires a voice search engine understand that there is a negative sentiment in the stock market and then search for news to analyze or reports that provide an explanation. After that, the voice search engine should generate the response in the form of a narrative. To accomplish that, the voice search engine needs to be able to understand the search criteria beyond simple keywords or phrases as well as to understand the fact that the user is expecting a simple answer.

I will continue with a list of unique features of voice search platforms in a future post.

CEO of IntoTheBlock, Chief Scientist at Invector Labs, I write The Sequence Newsletter, Guest lecturer at Columbia University, Angel Investor, Author, Speaker.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store