Russian-English system, Georgetown University

What it is about

A description of the Georgetown University Russian-English system as it ran on an IBM 360 mainframe. The paper gives a speed of about 9 seconds of CPU time per 1,000 words and a dictionary of about 50,000 entries, each with a Russian stem or split form, grammatical coding, a lexical number and an English part. Sentences are scanned repeatedly, from left to right and right to left, through four levels: word recognition, syntagmatic, syntactic and English synthesis. The main users were scientists at Oak Ridge, who used the output for information only, mostly without post-editing. The author says quality had not changed since 1964. A semantic level was planned, based on collocation and co-occurrence patterns counted from large corpora.

Ideas

• I1. Count collocations in the meta-world: for each pair of concepts that share a frame or a head-dependent link, count how often they co-occur in the Bible pairs, and let the encoder prefer the most frequent chunk in that context. This is the count-based form of the planned semantic level, and it can be measured as the share of tokens decided by a collocation count.
• I2. Ordered passes with stop rules: tools/roundtrip could run its tree-building passes in a fixed priority order and stop when the tree no longer changes. Log the number of passes per sentence, and check that two runs give the same trees.
• I3. Keep a dated table of residual per release, as the paper's flat quality since 1964 suggests the need: a system can stall for years without anyone noticing unless each release states its residual against the previous one.

For Mova

The paper is a system report with little method, so the useful part is the corpus-counted semantic level. It fits the frame and world layers (prag, world) and the count rule of the standard encoder. Any collocation count has to be measured on held-out verses first, and the stop rule of the passes has to be checked for determinism.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work.

@inproceedings{zarechnak1976russian,
  title = {{Russian-English system, Georgetown University}},
  author = {Zarechnak, Michael},
  booktitle = {Foreign Broadcast Information Service Seminar on Machine Translation},
  year = {1976},
  url = {https://aclanthology.org/1976.earlymt-1.13/}
}

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/zarechnak-1976-russian-english-system-georgetown-university/