{ "paper_id": "1984", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T01:11:21.502531Z" }, "title": "A BASIC SURVEY OF PRACTICAL MACHINE TRANSLATION", "authors": [ { "first": "Veronica", "middle": [], "last": "Lawson", "suffix": "", "affiliation": {}, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "1984", "_pdf_hash": "", "abstract": [], "body_text": [ { "text": "Machine translation (MT): what types of system are there, how well do they work, and will they improve?", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "This brief introduction to the different types of machine translation (MT) is intended as a kind of \"child's guide\" for those participants unfamiliar with the field. Its subject is not the linguistics of MT (which other papers will cover), but the practical applications of it, as viewed by a translator.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "If you recognise some of this, I apologise. Being a basic survey it is based, not surprisingly, on my earlier work, in particular my chapter on MT in \"The Translator's Handbook\" (8) .", "cite_spans": [ { "start": 178, "end": 181, "text": "(8)", "ref_id": "BIBREF7" } ], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "First, what is MT? For the purpose of this paper it is automatic translation, done by a computer with or without human assistance.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Definition", "sec_num": null }, { "text": "The machine offers translations of whole sentences.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Definition", "sec_num": null }, { "text": "It is not computerised term banks, word processors or other machine aids for translators (where the actual translating is done by a human).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Definition", "sec_num": null }, { "text": "(In reality, however, usage varies: \"MT\", \"machine aids\" and all their synonyms -machine-aided and computerassisted translation (MAT, CAT), mechanical and computer (-ised) translation etc. -may be applied to either concept or both, for both use machines to aid translation.)", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Definition", "sec_num": null }, { "text": "MT, then, need not be entirely automatic. Still less need it be fully automatic high quality translation (FAHQT), the idea rejected by the notorious ALPAC report (l) in 1966.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "How automatic?", "sec_num": null }, { "text": "Although that report effectively ended the US government's major MT research effort, work continued elsewhere, often on a small budget.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "How automatic?", "sec_num": null }, { "text": "(For MT's history, see Hutchins (4, 5) .) MT may have been less academically respectable, but its quality was improving, and the translation market was still expanding. Users increasingly found that MT,for all its limitations, could be worthwhile if used sensibly; and MT is now in practical use 1.", "cite_spans": [ { "start": 32, "end": 35, "text": "(4,", "ref_id": "BIBREF3" }, { "start": 36, "end": 38, "text": "5)", "ref_id": "BIBREF4" } ], "ref_spans": [], "eq_spans": [], "section": "How automatic?", "sec_num": null }, { "text": "in a number of organisations and for a variety of applications. It is not yet cheap enough for all translation services, nor will it ever (I believe) be good enough for all kinds of translation (far from it).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "How automatic?", "sec_num": null }, { "text": "The difference now is that people are prepared to acknowledge MT's limitations, and that the pressure for more and better MT, which before ALPAC came mainly from researchers, is now from users.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "How automatic?", "sec_num": null }, { "text": "So far, of course, machine translation has tended to be the preserve of large translation users such as government or intergovernmental agencies, multinationals or translation companies.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Who can obtain NT?", "sec_num": null }, { "text": "This situation, however, is beginning to change. Some of these users, like ITT in the UK, offer bureau facilities to smaller users or freelance translators, or alternatively the latter may form a consortium to buy a system.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Who can obtain NT?", "sec_num": null }, { "text": "NT may also be available on demand from a bureau (service centre): Weidner have one in Chicago, Control Data are to offer an ALPS bureau in Paris, and a Logos one exists in Switzerland.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Who can obtain NT?", "sec_num": null }, { "text": "With modern telecommunications, of course, the MT system need not be in-house: it can be run on a computer anywhere, and users can access it by telephone (line quality permitting).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Who can obtain NT?", "sec_num": null }, { "text": "Since editing needs less computer capacity than actual NT, therefore, a typical future pattern is likely to be a relatively large central computer for the actual MT and a number of distributed smaller computers for the editing.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Who can obtain NT?", "sec_num": null }, { "text": "Users wanting their own system can rent, lease or buy a program, with or without a computer.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "(The options vary with the NT supplier.)", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "In costing a scheme it is important to allow for all the initial costs.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "If the program is to run on the customer's existing hardware, for example, interface programs may have to be written at his expense; and production cannot even begin until after lengthy initial dictionary building.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "As for the running costs of an MT system, machine time (central processing unit time) makes up a relatively minor part of these.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "There is also the cost of input and output, handling, printing, and of course in most cases editing.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "The cost of a page of raw NT will probably be between \u00a31 and \u00a32, depending on the environment.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "Input will be cheaper, of course, if source texts are already in machinereadable form, and both input and output costs will be reduced if MT is integrated with office word processing.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "It has been suggested that a turnover of perhaps 2,000,000 words/year in a given language pair and broad subject area (e.g. chemistry) justifies the purchase of a high quality MT system on a mainframe computer.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "Such a system can multiply a translator's output by three to five, although probably only after a run-in period: performance in editing, like most professional skills, improves with experience and training.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "A simpler, less versatile system is likely to have a lower breakeven point, although it will not necessarily increase output by as much. Much depends, as usual, on the standard required and of course on the translator's previous output (probably between 300,000 and 900,000 words/year according to the time spent on research and non-translating duties).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "More information on MT costs can be found in Van Slype (12) .", "cite_spans": [ { "start": 55, "end": 59, "text": "(12)", "ref_id": "BIBREF11" } ], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "MT can be expected to come down in price as the price of computer storage falls and programs are tailored to smaller computers.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "Systran, designed for a mainframe computer, will also go on a minicomputer, Weidner and ALPS run on an IBM personal computer or Convergent Technologies micro as well as their usual minis, and ICL, for example, plan to offer MT on a micro shortly.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some points for costing", "sec_num": null }, { "text": "The MT field, like human translation, is very varied. Tables I and II show two ways of breaking it down, first by the type of system and then by the nature of any human intervention.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Kinds of machine translation", "sec_num": null }, { "text": "What place have these various kinds of MT in the translation market? Take a series of translation types such as the six Simpkin specifications (ll): literary, legal, publication, information, selective and gist.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "For the first -literary translation -MT appears to be a non-starter; it is doubtful whether even the most skilful editing can introduce the subtleties of rhythm on which literary style and nuance finally depend.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "For legal translation, too, MT may seem unpromising, although Systran can cope with some Euro-legalese if postedited, and even raw Systran may be useful to a legal user of patents (at least of chemical patents) who is prepared, as often happens, to do his own editing.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "For the purpose of (non-literary) publication, however, MT is already in use, although never raw unless its input is restricted (as with Meteo's weather forecasts). As for documents wanted for information, MT has been used for these for many years, in North American, Europe, the Soviet Union or the Far East.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "Full editing is not necessarily essential, for the expert reader's subject knowledge enables him to decode texts which a translator would regard as nonsense. As Lanna Castellano has said (3), the translator \"is far more concerned with the true quality of the product than is the user\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "Rapid postediting of only glaring errors suffices for many users, and even raw machine translation may be useful for information scanning.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "Raw MT can also assist in \"selective\" translation, by indicating which passages need an edited or human translation.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "\"Gist\", of course, is a different matter from pure translation, but presumably any MT good enough for information scanning can be expected to assist the person extracting the gist. What are the practical MT systems?", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "While there are numerous research systems, far fewer systems are actually used in \"real life\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "Only a small handful are widely marketed, the most successful so far being Systran and Weidner, followed by Logos and ALPS.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "(For a table of practical systems as of spring 1983, and also a selection of research projects, see Lawson (8) .)", "cite_spans": [ { "start": 100, "end": 110, "text": "Lawson (8)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "How much do MT dictionaries matter?", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "Many MT systems are available in more than one language pair.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "With luck the supplier will sell a given language pair more than once, but in this case the different customers are likely to produce quite different results.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "The normal practice is for all customers to receive periodic updates of the MT program, so that in one sense they all have the same system.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "Each customer's dictionaries, however, are tailor-made, with the result that, say, a European Commission Systran may perform better on administrative texts and worse on technical texts than a Canadian Systran (7) .", "cite_spans": [ { "start": 209, "end": 212, "text": "(7)", "ref_id": "BIBREF6" } ], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "Given the power of some MT dictionaries, with syntactic and semantic (10) coding and context rules, their effect on a program's performance is hardly surprising.", "cite_spans": [ { "start": 69, "end": 73, "text": "(10)", "ref_id": "BIBREF9" } ], "ref_spans": [], "eq_spans": [], "section": "What purposes?", "sec_num": null }, { "text": "The output of the many research-oriented MT projects, if any, is too limited, slow or expensive to be of commercial interest, but they dominate the MT literature. Even in the popular scientific press MT articles tend to be on these research projects, instead of on systems which are actually used.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "MT research", "sec_num": null }, { "text": "Academics must publish, whereas commercial companies must keep their knowhow secret (particularly software, which is notoriously hard to protect).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "MT research", "sec_num": null }, { "text": "The last three years have seen considerable changes, as machine translation has become more widely available. A notable difference is the greater acceptance in MT circles of the translator's role, partly because the European Commission has been demanding MT not only of a wider range of texts than any previous user, but of a higher quality.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Recent changes", "sec_num": null }, { "text": "It is now widely accepted that such MT requires the insight of the professional translator, as well as the skills of the computer scientist and linguistics expert.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Recent changes", "sec_num": null }, { "text": "Meanwhile hardware has become cheaper and commoner and started storing a higher proportion of documents.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Recent changes", "sec_num": null }, { "text": "The ratio of textual to numeric data stored on computers is estimated at 1:9 now, but is expected to reverse by the end of the decade.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Recent changes", "sec_num": null }, { "text": "The cliche 1 \"machine translation will never work\" is rarely heard now.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Recent changes", "sec_num": null }, { "text": "It obviously does work, within limits. But wall MT ever work perfectly? I believe that natural language is too subtle and changeable, too much alive, to machinetranslate perfectly -unless of course the system has been altered to cope with the particular text concerned. This kind of cheating or self-delusion was once found very tempting in MT circles, but has given way before the acknowledgment of MT's limitations. Users no longer necessarily expect human translation quality, and rigging has rather gone out of style.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Recent changes", "sec_num": null }, { "text": "\u00a9 Copyright Veronica Lawson 1983/4 6.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Languages and machines: computers in translation and linguistics. A report by the Automatic Language Processing Advisory Committee, Division of Behavioral Sciences", "authors": [ { "first": "", "middle": [], "last": "Alpac", "suffix": "" } ], "year": 1966, "venue": "National Academy of Sciences", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "ALPAC. Languages and machines: computers in translation and linguistics. A report by the Automatic Language Processing Advisory Committee, Division of Behavioral Sciences, National Academy of Sciences, National Research Council, Washington, D.C., 1966.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "The Posteditors' Experience\" (contributions by", "authors": [ { "first": "J", "middle": [ "A" ], "last": "Bachrach", "suffix": "" }, { "first": "R", "middle": [], "last": "Green", "suffix": "" }, { "first": "B", "middle": [], "last": "Lavorel", "suffix": "" }, { "first": "F", "middle": [], "last": "Rossi", "suffix": "" }, { "first": "S", "middle": [ "P" ], "last": "Sereda", "suffix": "" }, { "first": "D", "middle": [], "last": "Bostad", "suffix": "" } ], "year": null, "venue": "", "volume": "6", "issue": "", "pages": "97--136", "other_ids": {}, "num": null, "urls": [], "raw_text": "See session \"The Posteditors' Experience\" (contributions by BACHRACH, J.A., GREEN, R., LAVOREL, B., ROSSI, F., SEREDA, S.P. and BOSTAD, D.) in: ref. 6, 97-136.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "The practical tools employed", "authors": [ { "first": "L", "middle": [], "last": "Castellano", "suffix": "" } ], "year": null, "venue": "", "volume": "9", "issue": "", "pages": "47--79", "other_ids": {}, "num": null, "urls": [], "raw_text": "CASTELLANO, L. The practical tools employed. In: ref. 9, 47 -79.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Progress in documentation: machine translation and machine-aided translation", "authors": [ { "first": "W", "middle": [ "J" ], "last": "Hutchins", "suffix": "" } ], "year": 1978, "venue": "Journal of Documentation", "volume": "34", "issue": "2", "pages": "119--159", "other_ids": {}, "num": null, "urls": [], "raw_text": "HUTCHINS, W.J. Progress in documentation: machine translation and machine-aided translation. Journal of Documentation, 34(2), June 1978, 119-159.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "The evolution of machine translation systems", "authors": [ { "first": "W", "middle": [ "J" ], "last": "Hutchins", "suffix": "" } ], "year": null, "venue": "In: ref", "volume": "6", "issue": "", "pages": "21--37", "other_ids": {}, "num": null, "urls": [], "raw_text": "HUTCHINS, W.J. The evolution of machine translation systems. In: ref. 6, 21-37.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Practical Experience of Machine Translation", "authors": [ { "first": "V", "middle": [], "last": "Lawson", "suffix": "" } ], "year": 1981, "venue": "Proceedings of a conference", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "LAWSON, V., ed. Practical Experience of Machine Translation. Proceedings of a conference, London, 5/6 November 1981. Amsterdam: North-Holland, 1982.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Final report on EEC study contract TH-47", "authors": [ { "first": "V", "middle": [], "last": "Lawson", "suffix": "" } ], "year": 1982, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "LAWSON, V. Final report on EEC study contract TH-47, 1982. Methodology study on the application of the Systran computer- aided translation system to texts translated within the Commission of the European Communities.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Machine translation. In: ref. 9", "authors": [ { "first": "V", "middle": [], "last": "Lawson", "suffix": "" } ], "year": null, "venue": "", "volume": "", "issue": "", "pages": "81--88", "other_ids": {}, "num": null, "urls": [], "raw_text": "LAWSON, V. Machine translation. In: ref. 9, 81-88.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "The translator's handbook. London: Aslib", "authors": [ { "first": "C", "middle": [], "last": "Picken", "suffix": "" } ], "year": 1983, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "PICKEN, C, ed. The translator's handbook. London: Aslib,1983.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Theoretical options and practical limitations of using semantics to solve problems of natural language analysis and machine translation", "authors": [ { "first": "I", "middle": [ "M" ], "last": "Pigott", "suffix": "" } ], "year": 1979, "venue": "The Analysis of Meaning -Informatics 5", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "PIGOTT, I.M. Theoretical options and practical limitations of using semantics to solve problems of natural language analysis and machine translation. In: MacCafferty, M. and Gray, K., eds. The Analysis of Meaning -Informatics 5, 1979.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Translation specifications", "authors": [ { "first": "R", "middle": [ "E" ], "last": "Simpkin", "suffix": "" } ], "year": null, "venue": "", "volume": "9", "issue": "", "pages": "129--139", "other_ids": {}, "num": null, "urls": [], "raw_text": "SIMPKIN, R.E. Translation specifications. In: ref. 9, 129-139.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "Economic aspects of machine translation", "authors": [ { "first": "G", "middle": [], "last": "Van Slype", "suffix": "" } ], "year": null, "venue": "In: ref", "volume": "6", "issue": "", "pages": "79--93", "other_ids": {}, "num": null, "urls": [], "raw_text": "VAN SLYPE, G. Economic aspects of machine translation. In: ref. 6, 79-93.", "links": null } }, "ref_entries": { "FIGREF0": { "type_str": "figure", "num": null, "uris": null, "text": "the author's chapter on machine translation in C. PICKEN, ed. The Translator's Handbook London: Aslib, 1983 4." } } } }