Skip to main content
Fachgebiete
Automobil + Motoren Bauwesen + Immobilien Business IT + Informatik Elektrotechnik + Elektronik Energie + Nachhaltigkeit Finance + Banking Management + Führung Marketing + Vertrieb Maschinenbau + Werkstoffe Versicherung + Risiko
DE
EN
Bücher
Zeitschriften
Themenseiten
Organisationspsychologie Projektmanagement Marketing Smart Manufacturing
Weitere Formate
Podcasts Webinare Technik Webinare Wirtschaft Kongresse Veranstaltungskalender Awards
Jetzt Einzelzugang starten
Zugang für Unternehmen
Referenzkunden
SLX-Digitalkonferenz: Zukunftswerkstatt 2024

Mehr Umsatz mit Top-Kontakten

Am 7. Mai erfahren Sie, wie Vertriebsteams Leadgenerierung, Leadnurturing und Leadstrategien effizient steuern können.
Erweiterte Suche
Anmelden
nach oben
Erschienen in:
Introduction Kapitel lesen Erstes Kapitel lesen

2018 | OriginalPaper | Buchkapitel

2. Background

verfasst von : Bernard Scott

Erschienen in: Translation, Brains and the Computer

Verlag: Springer International Publishing

Einloggen

Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.

search-config
loading …

Abstract

This Chapter describes the exceptional circumstances that brought Logos Model MT into existence in 1969, and details the difficulties that confronted this pioneer development effort. Chief among the difficulties was the lack of proven models to guide the design and development of a workable MT system, causing Logos developers to turn for inspiration to assumptions about the processes taking place in human translation. Logos Model is contrasted in broad terms with statistical translation models, with which it shares certain resemblances. The eventual Logos Model translation process is then briefly described. The Chapter concludes with an overview of the basic assumptions about human translation processes that shaped Logos Model and that accounted for its early successes in the nascent MT world. The Chapter concludes with reflections about the nature and origin of language and grammar, all of which had a bearing on Logos Model design, development and performance. The advent of neural net MT is noted and the promise of this new development is briefly characterized.

Sie haben noch keine Lizenz? Dann Informieren Sie sich jetzt über unsere Produkte:

Springer Professional "Wirtschaft+Technik"

Online-Abonnement

Mit Springer Professional "Wirtschaft+Technik" erhalten Sie Zugriff auf:

  • über 102.000 Bücher
  • über 537 Zeitschriften

aus folgenden Fachgebieten:

  • Automobil + Motoren
  • Bauwesen + Immobilien
  • Business IT + Informatik
  • Elektrotechnik + Elektronik
  • Energie + Nachhaltigkeit
  • Finance + Banking
  • Management + Führung
  • Marketing + Vertrieb
  • Maschinenbau + Werkstoffe
  • Versicherung + Risiko

Jetzt Wissensvorsprung sichern!

Jetzt informieren

Springer Professional "Technik"

Online-Abonnement

Mit Springer Professional "Technik" erhalten Sie Zugriff auf:

  • über 67.000 Bücher
  • über 390 Zeitschriften

aus folgenden Fachgebieten:

  • Automobil + Motoren
  • Bauwesen + Immobilien
  • Business IT + Informatik
  • Elektrotechnik + Elektronik
  • Energie + Nachhaltigkeit
  • Maschinenbau + Werkstoffe




 

Jetzt Wissensvorsprung sichern!

Jetzt informieren

Springer Professional "Wirtschaft"

Online-Abonnement

Mit Springer Professional "Wirtschaft" erhalten Sie Zugriff auf:

  • über 67.000 Bücher
  • über 340 Zeitschriften

aus folgenden Fachgebieten:

  • Bauwesen + Immobilien
  • Business IT + Informatik
  • Finance + Banking
  • Management + Führung
  • Marketing + Vertrieb
  • Versicherung + Risiko




Jetzt Wissensvorsprung sichern!

Jetzt informieren
Vorheriges Kapitel Introduction
Nächstes Kapitel Language and Ambiguity: Psycholinguistic Perspectives
Anhänge
Nur mit Berechtigung zugänglich
Fußnoten
1
A report by the Automatic Language Advisory Committee published by the National Academy of Science in 1966.
 
2
During earlier service in the Air Force, the author worked as Vietnamese, Russian and French linguist.
 
3
Everett Pyatt, the early government advocate of our effort, was criticized for “chasing after fool’s gold” (personal communication). Pyatt later became Assistant Secretary of the Navy.
 
4
Personal communication to the author by Everett Pyatt, stating that this judgment appeared in the official (classified) Annual Report in 1973 of the then Director of Defense Research and Engineering (DDR&E), John S. Foster, Jr. Pyatt was attached to the Joint Chiefs of Staff at the time of this communication.
 
5
See Postscript 2-A for a brief note on the history of the commercial system.
 
6
Barreiro et al. (2014). In the Google Translate-Logos Model translation exercises described in this 2014 study, output for English-German favored Logos Model. But the case was just the opposite in the Romance language pairs.
 
7
Logos Corporation ceased operations at the turn of the century but a near equivalent copy of the commercial system continues to be available as OpenLogos, an open-source version of the commercial product, produced by DFKI in Germany. OpenLogos however has never undergone further development.
 
8
It is a misnomer to call Logos Model rule-based, although it is often associated with rule-based systems because virtually all other linguistic MT systems have been rule-based.
 
9
Evans (2014). This author states that cognitive linguists generally agree that the brain’s linguistic processes are pattern-based, not rule-driven.
 
10
Some neural MT models have begun to employ continuous processes that by-pass this initial alignment phase of SMT. See Kalchbrenner and Blunsom (2013).
 
11
Language models in Google Translate’s SMT system were derived from two trillion tokens of unlabeled monolingual text, yielding models comprising 300 million n-grams where n = 5 (Brants et al. 2007).
 
12
Koehn (2011, 305) suggests that for pairs like German-English, reordering requires annotating of words with part-of-speech tags and rules for their manipulation.
 
13
Google’s GNMT Translate now translates the main clause of (1) correctly, but renders the initial adjectival clause literally rather than idiomatically as its SMT system nicely did in 1(i). Microsoft’s Bing NMT Translator mistranslates the main clause of sentence (1): Seltsam, wie es scheinen mag, hat er nicht akzeptieren, die Promotion. Ironically, the earlier SMT version of Bing Translator (unshown) translated (1) correctly.
 
14
Every indication is that NMT technology is beginning to solve the morphology problem that has plagued SMT. For example, Google GNMT Translate now translates both (2) and (3) correctly. Bing NMT Translator also renders (2) and (3) correctly.
 
15
The character of this parse is linear rather than that of a traditional parse tree. This will be clarified in Chaps. 4 and 6 and in the discussion in Chap. 8 on Logos Model’s remote kinship with recursive, convolutional, deep neural nets.
 
16
Google’s new neural-net version of Google Translate (GNMT) now translates (7) with correct syntax: Il n’existe pas de nouvelles méthodes de renouvellement du crédit.
 
17
These semantically oriented rules are accessed in a Semantic Table called SEMTAB. Logos Model does not have provision for handling multiple senses of common nouns, one of the most difficult challenges facing linguistically based systems. (A linguist who worked on the European Community’s MT system in the 1990s told us they had written 700 rules to handle the transfers of a single source noun.) However, see Postscript 4-B in Chap. 4 for a conceptual Logos Model solution that was considered for this problem, one that was never implemented.
 
18
Most of the matters we address regarding ambiguity and complexity concern source analysis.
 
19
The case is otherwise of course whenever the decoding process becomes conscious and deliberate, as for example when the preconscious mind stumbles over a sentence and has virtually to parse it in order to untangle its import.
 
20
Of course, linguistic exposure in turn will condition thought itself, as Sapir and Whorf have argued.
 
21
See Postscript 2-B for discussion of mentalese.
 
22
No doubt considerations of felicity of style also entered importantly into the formulation of grammatical convention, but matters of style and felicity would be secondary to the more fundamental need to avoid misunderstanding.
 
23
See Haupt et al. (2008) for an obliquely related comprehension study of short German sentences with object-subject ambiguities and object-initial structures.
 
24
Given suitable context of course, the German sentence could possibly mean that this color suits my mother. Out of context, however, neither man nor machine would be expected to interpret it that way.
 
25
See Postscript 2-C for depiction of how (13) is processed by Logos Model to produce (14).
 
26
Curiously, the new neural net version of Bing Translator also translates (13) incorrectly.
 
27
This topic of verb types in Logos Model is graphically illustrated in Part II.
 
28
In Proceedings of MT Summit XV (2015), eds. Yaser Al-Onaizan and Will Lewis, papers on NMT dominate MT presentations for the first time.
 
29
Bengio (2009). LISA Lab’s NMT model is bidirectional, the first pass working from right to left, affording the second, left to right pass a degree of top-down intelligence about the entire sentence. See Proceedings of MT Summit XV ( 2015), eds. Yaser Al-Onaizan and Will Lewis.
 
30
Castilho et al. (2017) report that NMT outperformed SMT in six of 12 language pairs in formal translation exercises.
 
31
In Chap. 8 we relate Logos Model to this new development in neural net MT.
 
32
Scott (1990, 2000, 2003). Partly because of the requirements of corporate secrecy, and partly because of development pressures, nothing at all was published about Logos Model technology for the first twenty years of its existence, and only very little in the public domain after that. It is understandable, therefore, that the claims of this book may be difficult to recognize for readers familiar with the published history of MT.
 
33
See Chap. 9 Postscript for illustration of numeric representation in Logos Model.
 
Literatur
Al-Onaizan Y, Lewis W (eds) (2015) Proceedings of MT Summit XV
Barreiro A, Monti J, Orliac B, Arrieta K, Batista WF, Trancoso I (2014) Linguistic evaluation of support verb constructions by OpenLogos and Google Translate. In: Proceedings, language resources and evaluation. Reykjavik, Iceland, pp 26–31
Bengio Y (2009) Learning deep architectures for AI. Found Trends in Mach Learn 2(1):1–127CrossRef
Brants T, Popat A, Peng Xu, Och F, Dean J (2007) Large language models in machine translation. In: Proceedings of the 2007 joint EMNLP-CoNLL conference. Prague, Czech Republic.
Castilho S, Moorkens J, Gaspari F, Calisto I, Tinsley J, Way A (2017) Is neural machine translation the New State of the art? Prague Bull Math Linguist 108:109–120CrossRef
Evans V (2014) The language myth: why language is not an instinct. Cambridge University Press, Cambridge
Goldin-Meadow S, Mylander SC (1990) Beyond the input given. The child’s role in the acquisition of language. Language 66:323–355CrossRef
Haupt FS, Schlesewsky M, Roehm D, Friederici A, Bornkessel-Schlesewsky I (2008) The status of subject-object reanalysis in the language comprehension architecture. J Mem Lang 59(1):54–96CrossRef
Kalchbrenner N, Blunsom P (2013) Recurrent convolutional neural networks for discourse compositionality. In: Proceedings of the 2013 workshop on continuous vector space models and their compositionality. Sofia, Bulgaria, pp 119–126
Koehn P (2011) Statistical machine translation. Cambridge University Press, CambridgeMATH
Langacker RW (2008) Cognitive grammar: a basic introduction. Oxford University Press, New YorkCrossRef
Pinker S (1994) The language instinct. William Morrow and Co., New YorkCrossRef
Scott B (1990) Biological neural net for parsing long, complex sentences. Logos Corporation Publication
Metadaten
Titel
Background
verfasst von
Bernard Scott
Copyright-Jahr
2018
Buch
Translation, Brains and the Computer

Print ISBN: 978-3-319-76628-7

Electronic ISBN: 978-3-319-76629-4

Copyright-Jahr: 2018

DOI
https://doi.org/10.1007/978-3-319-76629-4_2