1.1. Historia de la traducción automatizada

History of Machine Translation

Introduction to Machine Translation

• The session begins with a greeting and a reminder about the homework on creating a timeline for machine translation, which is crucial in translation memory.

• The speaker introduces the topic of machine translation history as a precursor to the week's work.

Definition and Early Attempts

• Machine translation is defined as an area within computational linguistics that has significantly evolved since its inception in the 1950s.

• Early systems relied heavily on grammatical rules and predefined structures, leading to poor translations that often translated word-for-word.

Historical Context

• The concept of automated translation dates back to the 10th century during the scientific revolution, where mechanical or tabular systems were proposed for language conversion using rules and substitutions.

• These early methods involved coding words from one language into tables, allowing users to look up translations based on assigned codes rather than understanding the language itself.

Advancements in Technology

• By the 20th century, George Strony developed a paper tape storage device in 1933 that could find equivalent words across languages using punched cards as codes.

• Peter Smirnov Trojansky outlined three stages of machine translation: analysis by an editor familiar with the source language, transformation by machines into target language equivalents, and final adjustments by another editor fluent in the target language.

Key Developments Post-WWII

• A significant milestone occurred in 1949 with Weer’s memorandum outlining future prospects for machine translation, suggesting techniques like cryptography and statistical methods.

• In 1954, Leon Dostert collaborated with IBM for a public demonstration of machine translation translating Russian sentences into English using limited vocabulary and grammar rules.

Continued Research and Challenges

• From the 1950s to 1980s, numerous empirical studies emerged focusing on various approaches including bilingual dictionaries and statistical analyses.

• Notable contributions included Erwin Riffler's development of bilingual dictionaries for word-for-word translations and Anthony Odenger's extensive Russian-English dictionary compilation at Harvard.

History of Machine Translation

Development of Reversible Grammars

• The development of reversible grammars took place at the University of Texas and involved Sydney Lamp's strategic grammar model at Berkeley. Significant research also occurred in Europe and the Soviet Union, including interlingua models in Cambridge and Milan, as well as theoretical advancements like the Mediatex model in Moscow.

Evolution from Rule-Based to Statistical Models

• From the 1950s to the 1980s, studies aimed to refine automatic translation tools for efficiency. By the late 1980s, rule-based systems dominated machine translation research, relying on syntactic and lexical transfer rules. This marked a shift towards more complex linguistic frameworks.

• IBM's work in the late 1980s revolutionized machine translation by introducing purely statistical models that utilized parallel corpora. These models shifted focus from explicit linguistic rules to statistical analysis based on large datasets.

Limitations of Statistical Translators

• Statistical translators often produced incorrect translations due to reliance on frequency rather than accuracy; for example, translating "perro" (dog) as "cat" if it was statistically common despite being wrong. This highlighted a fundamental flaw in early statistical approaches to translation.

• Despite their limitations, these statistical translators could generate acceptable translations without explicit linguistic rules, paving the way for improved quality through advanced statistical techniques and minimal linguistic data usage.

Emergence of Example-Based Approaches

• In 1984, Makoto Nagao introduced example-based methods that reused previous translations by extracting equivalent text segments from parallel corpora. This approach proved effective because it relied on professional translator work rather than mere frequency counts.

Expansion in the 1990s

• The 1990s saw innovative research areas expand into multilingual text generation and spoken language translation systems like Castore and Verb. The use of machine translation surged within commercial agencies, government services, and multinational companies for technical translations at scale.

• Major systems such as Sistran, Logos, Metal, and Atlas became key players post-90s due to increased demand for translating large volumes of technical texts across various sectors including government and commerce.

Technological Advancements

• At the end of the 1990s, there was rapid growth in personal computer-based automatic translators alongside telecommunications networks; Japanese companies launched microcomputer systems while options like Global Link gained popularity in the U.S., integrating machine translation into electronic messaging platforms online.

• Innovations led to specialized systems designed for specific domains with controlled languages reflecting a diversification trend within machine translation tailored increasingly towards user needs over time.