16 May Discourse Pragmatics
Context data is critical for disambiguation of words that have many possible meanings. These meanings are almost always mediated by context, therefore understanding context must precede disambiguation. For instance, when standing in a ring appears in the context of a boxing match, ring refers to a physical object with corners. The same phrase, in a ceremonial context, may indicate the relative positions of the agents, and may involve no physical “ring” object. The ceremonial ring almost never has corners. The fact that a boxing ring has corners and a ceremonial ring does not may be expressed in a rule or an association. This rule or association is not really semantic, nor syntactic nor linguistically morphological. It’s a conceptual fact of the real world, and may reasonably be applied to the linguistic stratum of pragmatics which deals with context.
|Understanding Context Cross-Reference|
|Click on these Links to other posts and glossary/bibliography references|
|Prior Post||Next Post|
|Analyzing Semantics||Deixis and Context|
|Context semantic disambiguation||Huang 2007|
|understanding meaning pragmatics||Princeton Wikipedia Deixis|
|rule syntactic association||Aristotle|
Knowledge about quantifiers and world knowledge (such as “rings are round except in some sporting events like boxing and wrestling”) is pragmatic knowledge. Without pragmatic knowledge, accurate automatic translation is not possible. On the other hand, some forms of ambiguity and polysemy can be resolved using only morphological, syntactic, or semantic constraints. Because the computational cost of incorporating higher levels of knowledge increases, it has long been considered wise to have it available but use it only when necessary. What if we could reduce the cost dramatically? Would we be able to overcome the cost and use context all the time?
The relative inability of current translation programs to efficiently and accurately interpret and translate text that has not been pre-edited suggests that it is time to give pragmatics and context a more central role in analysis.
|boxing ring||wedding ring||ring around the collar||piston ring|
|familiar ring||Saturn's rings||rings true||o-ring|
A wedding ring has no corners, no beginning, and no end, but what about a crime ring?
One aspect of discourse pragmatics is deixis. In linguistics, deixis is the phenomenon, within discourse pragmatics, in which understanding the meaning of certain words and phrases in a spoken or written utterance requires contextual information. Words are described as deictic if their semantic interpretation is consistent across multiple contexts, but their denotational meaning varies depending on time and/or place or other contextual facts. Pronouns, and other words or phrases that always require contextual information to convey any meaning are inherently deictic.
Some levels of disambiguation require a combination of semantic and pragmatic knowledge. Consider, for example, quantifiers such as EACH and ALL:
“They watched each pitch thrown by all the pitchers.”
The words EACH and ALL can be disambiguated easily if the analyzer knows that pitcher is normally a singular AGENT. This singularity is a pragmatic constraint, so if we know that, in the context of baseball, only one pitcher throws pitches at a time, we will not erroneously interpret (or translate) the phrase to mean all the pitchers are throwing each of the pitches. The term “fast ball” is easily interpreted by anyone familiar with baseball, and is unambiguous throughout most of the western hemisphere and Japan, though there are occasions when the term is used in contexts other than baseball. In other parts of the world, the interpretation may be completely dissociated with baseball most of the time, and still be completely correct. My context based 3-Dimensional Grammar explicitly maintains context to solve this problem.
What about that runner rounding third and flying into the home stretch? When we have pragmatic knowledge about words and activities, we can differentiate between literal and figurative meanings of words. The likelihood of a major league baseball player taking to winged flight is certainly a long shot, yet he is the Agent of the verb flight. The idea of rounding third in some kind of aircraft (a semantic “Instrument”) also seems far-fetched. We know enough about the real world that when we hear that sentence, we automatically know what it means. If a group of people from diverse backgrounds can apply semantic and pragmatic constraints to correctly interpret, shouldn’t a computer be able to do the same?
Consider an earlier example about flying. Would knowing the context of the sentences surrounding these sentences help determine the best interpretations?
- I saw the Mississippi flying into Minneapolis. (one reasonable interpretation)
- I saw a condor flying over Ventura. (two reasonable interpretations)
- I saw my hang-gliding instructor flying into Denver. (multiple interpretations)
Would knowing the intended meaning affect the quality of a translation? Sometimes the words are enough to translate the source with the correct ambiguities. Often, though, it is not.
|Click below to look in each Understanding Context section|
|4||Perception and Cognition||5||Fuzzy Logic||6||Language and Dialog||7||Cybernetic Models|
|8||Apps and Processes||9||The End of Code||Glossary||Bibliography|