You are on page 1of 4



To understand something is to transform it from one representation into another, where this second representation has been chosen to correspond to a set of available actions that could be performed and where the mapping has been designed so that for each event, an appropriate action will be performed. If you say to an airline database system "I need to go to New York as soon possible," the system will have "understood" if it finds the first available plane to New York. If you say the same thing to your best friend, who knows that your family lives in New York, she will have "understood" she realizes that there may be a problem in your family and you may need some emotional support. As we talk about understanding, it is important to keep in mind that the success or failure of an "understanding" program can rarely be measured in an absolute sense & task to be performed. Understanding in various aspects: y y For people, understanding applies to inputs from all the senses. Computer understanding has applied primarily to images, speech, and typed language.

WHAT MAKES UNDERSTANDING HARD? There are four major factors that contribute to the difficulty of an understanding problem: 1. The complexity of the target representation into which the matching is being done 2. The type of the mapping: one-one, many-one, one-many, or many-many 3. The level of interaction of the components of the source representation 4. The presence of noise in the input to the understander Examples to illustrate factors of understanding: 1. Complexity of the Target Representation Suppose English sentences are being used for communication with a keyword-based data retrieval system. Then the sentence I want to read all about the last budget & result. would need to be translated into a representation such as (SEARCH KEYWORDS =BUDGET & RESULT) Extracting that information often requires the use of additional knowledge about the world described by the sentences. 2. Type of Mapping Understanding is the process of mapping an input from its original form to a more useful one. The simplest kind of mapping to deal with is one-to-one (i.e., each different statement maps to a single target representation that is different from that arising from any other statement). (a) one-to-one mapping: consider the language of arithmetic, expressions in many programming

or whatever). the two different patterns. Programming languages provide good examples of languages in which there is very little interaction among the components of an input. For example. Level of Interaction among Components In most interesting understanding contexts. A:B+C*D (b) many ±to-one In Many to one mappings free variantions are allowed ." A spectrogram shows how the sound energy is distributed over the auditory frequency range as a function of time. each input is composed of several components (lines. In fact. words. It shows a spectrogram produced by the beginning of the utterance "Alpha gets alpha minus beta. as the number of interactions increases.They are doing household work (SEARCH KEYWORDS= BUDGET & (d) Many ±to many there are many ways to say the same thing and a given statement may have many meanings." Even when we ignore the variability of the speech signal.No two people speak identically. such as Tell me all about thelast budget& result. each produced by the word "alpha.languages. := ** + . Otherwise. 3. The mapping process is the simplest if each component can be mapped without concern for the other components of the statement. so does the complexity of the mapping. symbols.----------------I like to see all reviews of budget and result.----------RESULT) I am interested in budget & result-----------------------Many-to-one mappings require that the understanding system know about all the ways that a target representation can be expressed in the source language.They are doing office work -----------. natural languages admit variability because of their richness.They are doing homework -----------. Another example of many-to-one mappings. one person does not always say a given word the same way. (c ) one-to-many It require a great deal of domain knowledge (in addition to the input itself) in order to make the correct choice among the available target representations. shows how changing one word of a statement requires only a single change to one node of the corresponding parse tree. In this example. An example of such a mapping (in which the input can be said to be ambiguous) is the following sentence: They are doing work ------------.

Look at one example of the use of this approach. We rarely have the opportunity to listen to each other against a background of silence. Consider the drawing shown in Fig. and the number of combinations of those components is enormous. UNDERSTANDING AS CONSTRAINT SATISFACTION Many understanding tasks appear impossibly complex. Although typed language is less susceptible to noise than is spoken language. or between objects and the background ‡ A Concave Edge²An edge between two faces that form an acute angle when viewed from outside the object . the image you will see i of the sign may be interfered with by many things. such as speech and image understanding. The same problem occurs in image understanding. There are two important steps in the use of constraints in problem-solving: 1. Assume either that you have been given this drawing as the input or that lower-level routines have already operated to extract these lines from an input photograph.3. such as your windshield wipers or the trees alongside the road. Unfortunately. Noise in Input Understanding is the process of "interpreting an input and assigning it meaning. noise is still a problem. In perceptual tasks. other things often interfere with the basic input before it reaches the understander. Thus we must take an input signal and separate the speech component from the background noise component in order to understand the speech. this problem is common. The next step in the analysis process is to determine the objects described by the lines. typing errors are common. But a closer analysis often reveals that many of the combinations cannot actually occur. The number of interpretations that can be assigned to individual components of an input is large. For example. If you look out of your car window in search of a particular store sign. These natural constraints can be exploited in the understanding process to reduce the complexity from unmanageable to tractable. Analyze the problem domain to determine what the constraints are. Solve the problem by applying a constraint satisfaction algorithm that uses control strategy. particularly if language is being used interactively to communicate with a computer system. Because of the complex environment in which understanding usually occurs. a similar problem involving local indeterminacy arises. We need first .COS In image-understanding problems as well. the Waltz algorithm for line identify each of the lines in the figure as representing either: ‡ An Obscuring Edge²A boundary between objects. 2. in many understanding situations the input to which meaning should be assigned is not always the input that is presented to the understander. 4.

+ Convex line Concave line Boundary line with interior to the right(down) Boundary line with interior to the right(up) ÷ . Determining the constraints We are trying to solve is how to recognize individual objects in a figure. This produces a set of four labels that can be attached to a given line. to handle these other edge types. To do that. For boundary lines. we also need to indicate a direction.‡ A Convex Edge²An edge between two faces that form an obtuse angle when viewed from outside the object Fig3: A Line Drawing For more complex figures.3 to show labellings. We use the conventions shown in Fig. such as cracks between coplanar faces and shadow edges between shadows and the background. would also be required. The approach we describe here has. in fact. we have to first label all the lines in the figure so that we know which ones correspond to boundaries between objects. other edge types. telling which side of the line corresponds to the object and which to the background.