Weitere Kapitel dieses Buchs durch Wischen aufrufen
This chapter, based largely on a paper originally written in 1976, deals with the problem of ambiguity as it relates to language acquisition and translation by both mind and machine. We define ambiguity as a linguistic situation capable of more than one interpretation and that obviously has bearing on the accuracy of translation. We enumerate five levels of ambiguity and describe the problem each of these ambiguity levels poses for a translation machine. We note, in contrast, that the mind is able to resolve these ambiguities virtually without thought, and we offer an explanation as to why this is so. We identify several psycholinguistic operations believed to be associated with the acquisition of a second language and that account for the progress of gifted learners, the ambiguities of language notwithstanding. We liken sentence analysis by a translation machine to human language acquisition, and proceed to show how the psycholinguistic factors involved in language acquisition, if simulated in the computer, enable the machine to cope with ambiguities with greater success than might otherwise be possible. Finally, we offer some classic examples of syntactic and semantic ambiguities that illustrate the disambiguating power of these psycholinguistic functions simulated in Logos Model.
Bitte loggen Sie sich ein, um Zugang zu diesem Inhalt zu erhalten
Sie möchten Zugang zu diesem Inhalt erhalten? Dann informieren Sie sich jetzt über unsere Produkte:
By equivalence here is meant literal translation for publication purposes, i.e., where accurate information transfer is of paramount consideration, the sort of translation we are concerned with. We are not concerned with translation for gisting purposes, and least of all with translation for purposes of edification, i.e., where style supersedes information transfer in the order of importance.
This view of syntax and semantics is consonant with that of Langacker and others in the school of construction grammar. See Schmid ( 1996, 353).
Ditransitive verbs, in more traditional parlance. We at Logos were prone at times to create our own vocabulary.
Of course with N3 could also be an adverbial phrase (e.g., with alacrity), which would change the interpretation of (10) entirely. However, the use of with in (10) for such purposes would confuse reader expectations and therefore would be highly unlikely if not bizarre (e.g., The school provided students with alacrity). This circumstance merely illustrates the strength of the claim that this particular group of verbs makes on the object of the preposition with as its own object, i.e., how strong the connection is between their semantics and their syntactic effect.
Interestingly, it’s the semantics of plumb line as “instrument” that causes the mind to drop the with respect to expectation that align had created regarding the preposition with, but this shift in interpretation only serves as further illustration of the effect semantics has on syntax.
Indeed, a teacher of a second language might use such a tree to depict, generically, the semantico-syntactic correspondences between the native language and the language being acquired. In the briefest of times, language students would begin to feel more comfortable with what they are being asked to learn.
Cited in Grishman et al. ( 1973, 433). Sager led the Linguistic String Parser effort at New York University’s Courant Institute until 1995. Her special focus was the treatment of syntactic ambiguity.
Sager was correct in thinking that word associations hold the key to disambiguation in (22), but only if the associations are between abstract, semantico-syntactic representations of words rather than between literal words.
This was born out by a number of random tests involving linguists applying for jobs at Logos Corporation.
It’s true of course that an expression like changes in cars produced by Ford suggests that cars and not changes are what Ford has produced. Sager’s point is certainly true that pragmatic associations like that between Ford and cars may (at times) be dispositive in ambiguous situations, but this is not necessarily so. For example, because of the strong semantico-syntactic association between produced and changes, the formulation of the above expression would more properly be changes in cars manufactured by Ford, just to avoid misreading.
A non-technical law, not to be confused with Chomsky’s formal law of subjacency, although clearly if spelled out technically, adjacency might not look very different. The difference of course has to do with whether the law denotes unique, innate linguistic behavior (Chomsky) or simply an intelligent, grammatical convention, much like the intelligence involved in not pointing to B if you wish to call attention to A.
As with previous examples, it is always possible to force a larger context that might allow such aberrant interpretation of (28), but if one could devise such a context, the exception would merely serve to prove the rule.
For our purposes here we have omitted enumerating SAL’s taxonomic set and subset codes that are also assigned to changes.
As sentence (24) passes through the analytical pipeline, it gets progressively rewritten until (24) in its entirety is reduced to NP (with changes as its head word).
The target action is not an actual component of the source pattern, but is linked to by the source action. In general, target linking is optional and a source rule whose chief purpose is analytical may not always imply a target action. Activation components for many different target languages can be linked to from a single source pattern-rule, making Logos Model multi-target: one source language analysis, multiple target language generations.
Notations for target language generation are created in Translation Step Four but generation of the literal translation occurs at the end of the translation process.
The French translation in (24)(iii) is taken from (Scott. 2003, 17) and was produced by a developmental version of Logos Model that never made it into a release prior to Logos Corporation shut down, possibly because the pattern-rule was too broad as written. In any case, (24)(iii) cannot be reproduced on OpenLogos. None of the other MT systems tested handled the gender of the participle in (24) correctly.
The commercial version of Logos Model, however, has been resurrected for use at KIIT University in Odisha, India in connection with its English-Hindi project. This came about as an outgrowth of earlier work with OpenLogos at the International Institute for Information Technology (IIIT) in Hyderabad, India, again in connection with English-Hindi MT.
It should be noted that output from OpenLogos on rare occasions is unable to reproduce output of the final release of the commercial system in 2000. The reason for this output difference is unknown.
Zurück zum Zitat Chomsky N (1957) Syntactic structures. Mouton, The Hague/Paris MATH Chomsky N (1957) Syntactic structures. Mouton, The Hague/Paris MATH
Zurück zum Zitat Chomsky N (2006) Language and mind. Cambridge University Press, Cambridge CrossRef Chomsky N (2006) Language and mind. Cambridge University Press, Cambridge CrossRef
Zurück zum Zitat Grishman R, Sager N, Raze C, Bookchin B (1973) The linguistic string parser. In: Proceedings of the national computer conference, pp 427–434 Grishman R, Sager N, Raze C, Bookchin B (1973) The linguistic string parser. In: Proceedings of the national computer conference, pp 427–434
Zurück zum Zitat Koehn P (2011) Statistical machine translation. Cambridge University Press, Cambridge MATH Koehn P (2011) Statistical machine translation. Cambridge University Press, Cambridge MATH
Zurück zum Zitat Schmid P (1996) Clausal constituent shifts: a study in cognitive grammar and machine translation. Ph.D. Dissertation, Georgetown University. UMI Dissertation Services Schmid P (1996) Clausal constituent shifts: a study in cognitive grammar and machine translation. Ph.D. Dissertation, Georgetown University. UMI Dissertation Services
Zurück zum Zitat Scott B (2003) Logos model: an historical perspective. Mach Transl 18(1):1–72 MathSciNetCrossRef Scott B (2003) Logos model: an historical perspective. Mach Transl 18(1):1–72 MathSciNetCrossRef
- Language and Ambiguity: Psycholinguistic Perspectives
- Chapter 3
Neuer Inhalt/© ITandMEDIA