Wolfram Alpha lets Siri answer product-shopping questions

Background

Siri, the voice assistant demonstrated with Apple's iPhone 4S, works in stages: speech is transcribed into text, the text is interpreted as a request, and the request is then handed to some source that can supply an answer. One such source is the Wolfram Alpha computational knowledge engine. At the iPhone 4S launch on 4 October 2011, Apple's Scott Forstall demonstrated Siri by asking how many days remained until Christmas, and the answer Siri gave came from Wolfram Alpha ^[2]. The choice is telling: Wolfram Alpha does not return a list of web pages, it computes a result from curated data, so a question about a date can be answered on the spot.

For a user, saying a sentence is easier than filling in filter boxes on a web page. For a system, it is harder: one spoken sentence often carries several conditions at once, and each of them has to be separated out and matched to a field in a database before it becomes a usable query.

What happened

On 19 December 2011, AppleInsider reported that Wolfram Alpha had enabled Siri users to shop for electronics by voice ^[1]. Wolfram Alpha said it was drawing on data from Best Buy's public APIs, so that a request could be run as a query against the retailer's database of 35,000 consumer products ^[1]. An API, or application programming interface, is an agreed way for one piece of software to ask another for data; a public API lets an outside service query a company's catalogue by rule instead of copying the whole catalogue over.

What a user could ask

According to AppleInsider, a user could search for a broad product category or for a specific model, and get back product features, prices, availability, user ratings and comparisons ^[1]. Wolfram Alpha's own illustration was a request for the lightest digital camera under $100 with resolution greater than 12 megapixels ^[1]. That example shows what structured data buys: the question stacks a price ceiling, a physical property (weight) and a technical threshold (resolution) on top of one another. A keyword search struggles with that; a database query can filter on all three conditions at once. The Best Buy data was available only to users in the United States ^[1].

What it shows

The same division of labour appears in both examples. Listening and interpreting belong to the assistant, while the domain knowledge comes from an outside engine: the Christmas countdown came from Wolfram Alpha ^[2], and the camera question was answered against a retailer's product data reached through its APIs ^[1]. It also shows what a shopping question needs that a fact question does not. A countdown to a fixed date is settled knowledge; a price and a stock status change, so the query has to land on a catalogue that is kept current rather than on a fixed passage of text.

The reach of the feature was part of the story. As reported at the time, the data behind it was one retailer's database of 35,000 consumer products, and it was open only to users in the United States ^[1]. The assistant could answer a fairly precise shopping question — lightest camera, budget ceiling, minimum resolution — but the answers came from a single source and a single market.