CS789, Spring 2007

Lecture notes, Week 9

Linguistic Interaction


We like language for interfaces because it's familiar to so many users. We can take advantage of an important communication mechanism that users already know how to use. Language has some properties of which we must be aware

At the current state of the art interfaces can emit language, but they can't accept it very well.

Spoken versus written language

Spoken and written language are different one from another. With spoken language

Written language is different. The reader has

If we are planning to use spoken language extensively in an interface it is obviously useful to give it some of the attractive properties of written language.

Speech acts

Language can be described in many different ways. The important one for us is what it does. In technical terms, this is the performative use of language, individual parts of which are called speech acts (J. L. Austin). Examples:

Our linguistic interaction with interfaces is mainly performative.

The philosopher H. P. Grice did a lot of work trying to figure out the properties of speech acts. This turns out to be very hard. Why? Because there's really a lot of understood content everywhere in language. He formulated the Cooperative Principle, which says that each party in a conversation assumes that the conversation itself is a cooperative activity, and that each party assumes the other parties know the conversation to be cooperative, and so on. Cooperation is based on the idea that a dialogue has a purpose. Thus, acts are formulated in language -- words, sentences, paragraphs -- in the context of mutual agreement with respect to the purpose of the dialogue. To be sure, purpose evolves during a conversation. But Grice was able to identify some abstract properties embodied in all cooperative language acts, regardless of actual purpose.

How are these abstract properties used in practice?

Grice unpacked the Cooperative Principle into four maxims of conversation, which can be used as design rules when making language for speech acts.

  1. Maxim of Quantity -- All the necessary information is given; no unnecessary information is given.
  2. Maxim of Quality -- Information given is true, not false or misleading.
  3. Maxim of Relation -- Information given is relevant to the goal of the conversation.
  4. Maxim of Manner -- The communication is clear, brief and orderly; ambiguity and obscurity is avoided.

Examples of failures:

  1. Maxim of Quantity -- Extra information is given. "On this trip the bus has a driver."
  2. Maxim of Quality -- Some information is obviously inconsistent. "Leaves Kitchener at 8.00am; arrives Guelph at 7.30am."
  3. Maxim of Relation -- Some of the information seems to be irrelevant. "There is a good view of the CN tower on this trip."
  4. Maxim of Manner -- Ambiguity and obscurity is present. "Leaves Kitchener twenty-three minutes after the mayor starts his speech."
Here are a few examples of meta-processing
  1. Maxim of Quantity -- "Tickets are required for this trip." Where do I find out about trips that don't need tickets?
  2. Maxim of Quality -- "Express trip, stopping at Guelph." How much more of the message is not correct? or possiblly, Which part of this message is not correct?
  3. Maxim of Relation -- "Beautiful turtles throng the aisles." Where do I find the table of contents?
  4. Maxim of Manner -- "Runs on days with `r' in months when oysters are in season." Is there somebody somewhere who speaks English?

Here are a few other implications of these maxims

  1. Maxim of Quantity
  2. Maxim of Quality
  3. Maxim of Relation
  4. Maxim of Manner
There are two other principles that are sometimes tacked on with Grice's principles.

Special Considerations when One Party is a Computer

The above principles apply equally when both parties are human, and also when one party is a computer. There are, in addition, a few extra principles that can help to overcome the limitations of a computer interlocutor.

  1. Assymmetry -- help the user to understand the assymmetry of comprehension/expression that exists between the user and the computer.
  2. Background knowledge -- because interfaces cannot easily infer the background knowledge a user has, make the interface work with as wide a variety of background assumptions as possible.
  3. Repair and clarification -- repair requires 'meta-communication', communication about the current state of communication.

Grice's Maxims play a role in all dialogue design. Why are they particularly important for linguistic dialogues? Consider the serial nature of language.

Return to: