The Question Parser Eduction Grammar

The Fact Bank Question Parser is an IDOL Eduction grammar designed for question answering, which the Answer Server internal Eduction module uses to parse questions. This special grammar file defines many different forms that questions can take, and extracts the entities, properties, and qualifiers in the questions.

In general, an entity is the topic of the question, or a topic in your Fact Store data. Entities have properties, which define pieces of information that you might want to find, or which you might want to use to find a particular entity. A qualifier is a piece of information that modifies the property.

For example, in the question What is the population of the USA in 1850, you might define:

  • USA to be the entity.
  • population to be the property.
  • 1850 to be a qualifier.

Many questions can be interpreted in more than one way, depending on how you set up your data.

The Fact Bank grammar can find the entities, properties, and qualifiers in a variety of different question formations.

When the Question Parser processes a question, the grammar might find several valid interpretations. In this case, the Question Parser returns all the options to Answer Server, which attempts to find the associated entities, properties, and qualifiers in your coding files, and then in your data.

The values that match in your Fact Store at this stage depend on how you have set up and stored your data.

NOTE: For time and entity type answers, Answer Server merges duplicate answers when multiple interpretations return the same value.

The examples in the sections below demonstrate many of the different forms of questions in English that the Question Parser can detect. The questions are examples, and the list is not exhaustive. In addition, some of the forms are not mutually exclusive (that is, Question Parser might detect both forms). In these cases, the correct form depends on your data.

Processors

The Q&A grammar also detects a number of processors, which affect the question in a similar way to qualifiers, but which might require further calculation on the data. The following processors are supported (processors listed on the same line are equivalent):

  • min, minimum
  • max, maximum
  • mean, average
  • total, sum
  • first, oldest
  • last, latest

Example Questions

The following section lists several example questions in forms that the Question Parser can parse, and which Answer Server can use to retrieve the relevant facts from the Fact Store.

In the questions, entities are specified in italics, properties are specified in bold, and qualifiers are underlined.

Questions to Find a Property Value

The following questions show forms where the Question Parser detects an entity and property name, and any associated qualifiers. For these questions, Answer Server attempts to find the property value in the Fact Store.

  • What is the population of France?

  • What is France's population?

  • Can you tell me the French population?

  • The population of France is what?

  • How many people live in France?

    This question also matches the property number of people (the grammar automatically constructs this property because of the how many in the question).

  • What was the population of France in 2010?

  • What is the female population of France?

  • What is the non-female population of France?

    In this example, the qualifier female is negated, so Answer Server finds facts that do not have this qualifier.

  • What is the population of France that is female?

  • What is the population of France that isn't female?

    In this example, the qualifier female is negated, so Answer Server finds facts that do not have this qualifier.

  • What did Dickens write in 1837?

  • Who discovered America in 1492?

    This question also matches the property discovered by (the grammar automatically constructs this property because of the who in the question).

  • What is the France population above 60000000?

Questions to Find Processed Property Value

The following questions show forms where the Question Parser detects an entity and property name, and a processor that qualifies the value that the question requests. For these questions, Answer Server attempts to find the property value in the Fact Store.

Processors are marked in monospace font.

  • What is the latest population of Russia?

  • What is the maximum male population of Russia?

  • When was George Washington first elected?

  • What are the total electoral votes of George Washington?

Questions to Find the Name of an Entity

The following questions show forms where the Question Parser detects a property and one or more property values. For these questions, Answer Server attempts to find the name of an entity in the Fact Store.

Property values are marked in bold italic.

  • Where is Paris the capital of?

  • Whose capital is Paris?

  • Which country has a capital that is Paris?

  • Who was president in 1810?

  • Who was president of America in 1810?

  • Who are the presidents of America who are fictional?

    In this example, the fictional qualifier modifies the property presidents, which has the value America.

  • Who are the presidents of America who are male?

    In this example, male is treated as the value of a separate unnamed property of the entity that the question requests.

  • What country borders France and Spain?

  • What country borders Portugal?

  • What products are there with interest rates above 0.3%?

  • What products are there with interest rates between 0.1 and 0.3%?

Questions to Find the Value of a Qualifier

The following questions show forms where the Question Parser detects an entity, a property, and a property value. For these questions, Answer Server attempts to find the value of a qualifier that matches these values.

  • When did the USA have George Washington as president?

    This question matches a special point in time qualifier.

  • When did George Washington become president?

    This question matches the special position held property, and a special start time qualifier.