How Siri Works and Why It Matters for Local
This smart post from Simon Pensen first appeared a few weeks back on Search Engine Watch, though it is new to me. In it, Pensen makes the case that Google and Apple are in a face-off for the future of search — a future that looks sort of like Tim Berners-Lee’s vision of the semantic web, with some notable differences.
Devoting most of his analysis to what he sees as the seismic shift behind Google’s Penguin update and the development of the Knowledge Graph, Pensen argues that Google is attempting to compete with Siri on the semantic interpretation of search queries, and more specifically with Wolfram|Alpha, the “computational knowledge engine” that powers Siri’s responses to your questions about the population of Swaziland and the average rainfall in Zürich.
Pensen makes two interesting, though I think somewhat hasty, assumptions: first, that Wolfram|Alpha is analogous to Google’s emerging vision of a semantic graph for search; second, that Wolfram|Alpha is fundamental to the way Siri processes search queries. As I say, I don’t think either of these assumptions is precisely correct, but they do spark one to think about Wolfram|Alpha’s potential uses for all kinds of things, not least of which is local search.
Siri is usually described as a voice recognition engine tied to a set of web services and linked apps. The set includes Wolfram|Alpha for general knowledge questions; Yelp/Apple Maps for local search; Yahoo for sports scores, weather, and stocks; Safari for web search; the iPhone’s Clock, Reminders, Calendar, and Calculator apps; and so on. Siri also has the ability to communicate via web service to its own proprietary bank of responses for certain questions. Ask it for the average rainfall in Zürich and Wolfram|Alpha takes over, but ask for the airspeed velocity of an unladen swallow and Siri gets the joke.
So it’s pretty clear that Siri’s interpreter can examine a spoken query for syntax and keywords in order to trigger what it thinks is the most relevant web service. Often when Siri gets it wrong, this is because it has made a mistake about which service to call. In my experience as an iPhone user, Siri is somewhat over-eager to assume you want local businesses when you say a word that sounds like a product or service category. No doubt this is because Apple wants you to be able to say “restaurant” and get local results immediately.
In the few milliseconds it takes for a query to reach Wolfram|Alpha, Siri has already determined it to be a query for general knowledge, and at that point, we are dealing with a closed universe of information, a “computable knowledge space” which Wolfram|Alpha’s website describes as “curated” (echoes of pre-Google web search!). Wolfram|Alpha is something like an interactive Encyclopedia Britannica that turns to just the right page when you ask it a question.
The conclusion would appear to be that Wolfram|Alpha’s own interpreter has basically no interaction with the larger web, except in the very minimal sense that it competes to be the chosen web service during Siri’s initial analysis. Even then, I think I’m seeing web search take a backseat to all other query types; when Siri can’t come up with a better answer, it defaults to saying “Would you like me to search the web?” I certainly don’t get the sense that Siri is comparing Wolfram|Alpha’s content against that of general web search in order to decide which to display.
Hopefully this clarifies where the differences lie between Siri and Google’s Knowledge Graph. Though you may not know it by name, Knowledge Graph has probably become familiar to you since its debut in the spring of this year. It’s what happens when you search for Leonardo da Vinci and see his picture and basic facts about his life from Wikipedia, along with suggestions for related searches, at the top of the Google organic page. For mobile, Knowledge Graph is helping to power Google’s voice assistant for Android and is the key to why Android has a slight advantage over Siri in its ability to deliver fast responses to search and knowledge questions. Putting it somewhat crudely, Knowledge Graph is Google’s attempt to overlay a semantic layer on organic search and make it behave in the context of the larger web like Wolfram|Alpha does within its self-contained “knowledge space.”
Neither one of these things is the same as the Berners-Lee vision of the semantic web, which relies on the widespread adoption of schemas. Both may be looking at the problem in a more realistically scalable way, however. Where Wolfram|Alpha has an edge is in its lexical and syntactic analysis. Whereas Google can collate popular and authoritative search results around a term like “Leonardo da Vinci,” it has limited capacity today to interpret more complex phrases, something Wolfram|Alpha handles beautifully.
Here’s a quick measure of the difference. Google and Wolfram|Alpha answer the following questions equally well:
- What is the population of Swaziland?
- What is the capital of Swaziland?
Only Wolfram|Alpha offers answers to these:
- What is the average rainfall of Swaziland?
- What is the population of the capital city of Swaziland?
Google search results contain the answers to the latter questions, to be sure, but they are not captured by the Knowledge Graph, so Google doesn’t know it has the answer.
It’s a bolder proposition to turn syntactic interpretation loose on the web as a whole (though it should be noted that Ask.com has cornered a segment of this market). One wonders what would happen if the Wolfram|Alpha interpreter along with its curated body of knowledge were somehow used as a structuring element for the larger web. More down to earth of course would be the application of natural language processing to a closed system like that of a local search directory. The Wolfram|Alpha interpreter could surely make sense of phrases like “Find a dry cleaner within 10 miles that offers 24-hour service,” or “Find electronics stores that have flatscreen TVs on sale.” Much of that structured data is already being compiled by Apple Maps; it would only be a matter of training the interpreter to match complex queries to business profile details.
The announcement that Apple has hired Bill Stasior, founder of Amazon’s A9 search platform, as the new head of its Siri team speaks to the company’s seriousness about improving its search capabilities. Amazon search is, of course, the same type of closed system as Wolfram|Alpha. Taken by itself, Wolfram|Alpha is one of the stronger components of the Siri offering today. Why not leverage that strength to bring semantic search to local?
Damian Rollison is VP of Product and Technology at Universal Business Listing, a company dedicated to promoting online visibility for local businesses. Damian holds degrees from UC Berkeley and the University of Virginia, where he worked at the Institute for Advanced Technology in the Humanities. You can connect with him on Twitter.