--- title: "The grammatical tagging of unrestricted English text (1984)" description: "The paper describes how the Lancaster team, working with colleagues in Oslo and… 77% accuracy. 6 references, cited by 1 paper here." authors: ["Roger Garside", "Geoffrey Leach", "Eric Atwell"] year: 1984 venue: "Proceedings of the International Conference on Methodology and Techniques of Machine Translation: Processing from words to language" source: "https://aclanthology.org/1984.bcs-1.20/" licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/garside-1984-grammatical-tagging-unrestricted-english-text/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/garside-1984-grammatical-tagging-unrestricted-english-text/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [1984](https://md.research.elamur.ai/free/en/papers/1984/) / The grammatical tagging of unrestricted English text # The grammatical tagging of unrestricted English text **Authors:** [Roger Garside](https://md.research.elamur.ai/free/en/authors/roger-garside/), Geoffrey Leach and [Eric Atwell](https://md.research.elamur.ai/free/en/authors/eric-atwell/) · **Year:** 1984 · **Venue:** [Proceedings of the International Conference on Methodology and Techniques of Machine Translation: Processing from words to language](https://md.research.elamur.ai/free/en/venues/bcs/) ([International Conference on the Challenge of Arabic for NLP/MT 1984](https://md.research.elamur.ai/free/en/venues/bcs/1984/)) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/) **Links:** [Publisher's page](https://aclanthology.org/1984.bcs-1.20/) · [PDF](https://aclanthology.org/1984.bcs-1.20.pdf) **Citations:** 1 in our library, of the 149,439 papers whose bibliographies we read ## Summary The paper describes how the Lancaster team, working with colleagues in Oslo and Bergen, automatically assigned part-of-speech tags to the LOB corpus. LOB is a million-word collection of written British English built to parallel the Brown corpus. An earlier automatic tagging of Brown had reached only about 77% accuracy. Two things make the task hard: English has many homographs, and its vocabulary is open-ended. The system is a pipeline: - A "verticalising" step puts one word or punctuation mark on each record, splits enclitics and normalises letter case. - A light manual pre-edit follows. - WORDTAG proposes candidate tags for each word in isolation, using a small wordlist plus a suffix list. - IDIOMTAG adjusts tags for multi-word phrases. - CHAINPROBS chooses among the remaining candidates using probabilities of tag sequences learned from tagged Brown data. - Two passes of manual post-editing finish the job. The automatic stages pick the right tag roughly 96–97% of the time. This made probabilistic tagging practical for unrestricted text. The tagged corpus can then be used to train the next round of tagging. ## Key points - LOB has 500 text extracts of about 2,000 words each, grouped into 15 genre categories. It mirrors the Brown corpus, and the authors defined a refined 134-tag LOB tagset. One refinement splits the Brown proper-noun tag into NP, NPL, NPT and NNP. - WORDTAG relies on a wordlist of about 7,200 entries and roughly 700 suffix rules, tried from the most to the least specific ending. For a typical batch of records: - about 20% arrive already tagged, - 65% are resolved by the wordlist, - 9% are resolved by the suffix list, - the rest go to fallbacks such as stripping a final -s or a default of "noun, verb or adjective". - After WORDTAG, about 40% of units still carry more than one tag. CHAINPROBS ranks these candidates with probabilities and discards alternatives when it is confident. Weighted tag triples handle cases such as an intervening adverb. - IDIOMTAG covers about 150 phrases (e.g. "as … as"). It introduces "ditto" tags so that a multi-word unit like "according to" gets a single grammatical tag. - Overall automatic accuracy is about 96–97%. A Bergen trial that replaced manual pre-editing with extra automatic processing still scored above 90%. ## Interesting ideas - The tagger is split into separate stages: lexical candidates, phrase overrides, then contextual choice. Each stage records how it reached its decision, so errors can be traced to their source afterwards. - Suffix rules are matched from longest to shortest ending, and every word that would be mistagged by its suffix goes into the wordlist instead. The two lists are therefore built together as one system. - The disambiguator does not force a single answer. It keeps a ranked list of tags with probabilities and only removes alternatives above a confidence threshold, leaving unclear cases for a later checker. - A "ditto" tag lets several orthographic words act as one syntactic unit, such as a complex preposition or a conjunction like "so that". - Bootstrapping is incremental: each newly tagged and corrected corpus supplies the probability tables for tagging the next one. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (tokens): The en tagger could log, for every token, whether its tags came from the lexicon, a suffix rule, a multi-word rule or contextual scoring, which keeps each decision explainable and makes error audits easy. - **I2** (tokens): Unknown-word guessing in crate en could use a longest-suffix cascade with an exception lexicon that is checked first, giving a small and transparent model for out-of-vocabulary words. - **I3** (learn): Silver-data generation could keep low-confidence tokens as ranked alternatives and pass them to an LLM teacher tier, instead of committing early and spreading errors. - **I4** (tree): Multi-word expressions like "according to" could be marked before parsing and mapped onto UD fixed relations, so the parser does not have to rediscover them. - **I5** (learn): Mova's repaired silver treebanks could be fed back as the statistics source for the next training round, with negative-control evaluation at each step to check the loop stays honest. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* 📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works 🧭 Topics (1): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,263 works ## Cite Cite as: Roger Garside, Geoffrey Leach and Eric Atwell. 1984. The grammatical tagging of unrestricted English text. Proceedings of the International Conference on Methodology and Techniques of Machine Translation: Processing from words to language. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{garside1984grammatical, title = {{The grammatical tagging of unrestricted English text}}, author = {Roger Garside, Geoffrey Leach and Eric Atwell}, booktitle = {Proceedings of the International Conference on Methodology and Techniques of Machine Translation: Processing from words to language}, year = {1984}, url = {https://aclanthology.org/1984.bcs-1.20/} } ``` - [← previous English and the Class of Context-Free Languages](https://md.research.elamur.ai/free/en/papers/postal-1984-english-class-context-free-languages/) - [↑ back to papers of 1984](https://md.research.elamur.ai/free/en/papers/1984/) ## References 6 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Francis, W.N., and Kučera, H. (1979) Manual of Information to Accompany a Standard Corpus of Present-Day Edited American English, for Use with Digital Computers, Providence, R.I.: Department of Linguistics, Brown University. [find](https://scholar.google.com/scholar?q=Francis%2C+W%2EN%2E%2C+and+Ku%C4%8Dera%2C+H%2E+%281979%29+Manual+of+Information+to+Accompany+a+Standard+Corpus+of+Present%2DDay+Edited+American+English%2C+for+Use+with+Digital+Computers%2C+Providence%2C+R%2EI%2E%3A+Department+of+Lingui) 2. Greene, B.B. and Rubin, G.M. (1971). Automatic Grammatical Tagging of English, Providence, R.I.: Department of Linguistics, Brown University. [find](https://scholar.google.com/scholar?q=Greene%2C+B%2EB%2E+and+Rubin%2C+G%2EM%2E+%281971%29%2E+Automatic+Grammatical+Tagging+of+English%2C+Providence%2C+R%2EI%2E%3A+Department+of+Linguistics%2C+Brown+University%2E) 3. Hofland, K. (1983) in 'Seminar on the Use of Computers in English Language Research', ICAME News 7, 1-12. [find](https://scholar.google.com/scholar?q=Hofland%2C+K%2E+%281983%29+in+%27Seminar+on+the+Use+of+Computers+in+English+Language+Research%27%2C+ICAME+News+7%2C+1%2D12%2E) 4. Johansson, S. and Jahr, M-C. (1982). 'Grammatical Tagging of the LOB Corpus: Predicting Word Class from Word Endings' in S.Johansson, ed. Computer Corpora in English Language Research, Bergen: Norwegian Computing Centre for the Humanities. 118-46. [find](https://scholar.google.com/scholar?q=Johansson%2C+S%2E+and+Jahr%2C+M%2DC%2E+%281982%29%2E+%27Grammatical+Tagging+of+the+LOB+Corpus%3A+Predicting+Word+Class+from+Word+Endings%27+in+S%2EJohansson%2C+ed%2E+Computer+Corpora+in+English+Language+Research%2C+Bergen%3A+Norwegi) 5. Johansson, S., Leech, G.N. and Goodluck, H. (1978) Manual of Information to Accompany the Lancaster-Oslo/Bergen Corpus of British English, for use with Digital Computers, Oslo: Department of English, University of Oslo. [find](https://scholar.google.com/scholar?q=Johansson%2C+S%2E%2C+Leech%2C+G%2EN%2E+and+Goodluck%2C+H%2E+%281978%29+Manual+of+Information+to+Accompany+the+Lancaster%2DOslo%2FBergen+Corpus+of+British+English%2C+for+use+with+Digital+Computers%2C+Oslo%3A+Department+of+English%2C+) 6. Marshall, I. (1984) 'Choice of Grammatical Word-Class without Global Syntactic Analysis for Tagging Words in the LOB Corpus', Computers and the Humanities (to appear). [find](https://scholar.google.com/scholar?q=Marshall%2C+I%2E+%281984%29+%27Choice+of+Grammatical+Word%2DClass+without+Global+Syntactic+Analysis+for+Tagging+Words+in+the+LOB+Corpus%27%2C+Computers+and+the+Humanities+%28to+appear%29%2E) *Updated 29 October 2026*