Language and Computers

Citation

Glass, L., Dickinson, M., Brew, C., & Meurers, D. (2024). Language and Computers (2nd ed.; Textbooks in Language Sciences 14). Language Science Press. Free official edition: https://langsci-press.org/catalog/book/454

Chapter Summary

Chapter 1: Encoding language on computers

The book opens by asking what language is, using Hockett's design features, and by separating language from writing. It surveys writing systems (alphabets, abjads, abugidas, syllabaries and logographic systems), how computers represent text with binary numbers, ASCII and Unicode/UTF-8, and how speech is captured and analysed acoustically, with the International Phonetic Alphabet and spectrograms. Short sections introduce automatic speech recognition and text-to-speech before the book sets speech aside to focus on text. "Under the Hood" boxes cover emoji, online gig platforms and reading a spectrogram.

Chapter 2: Writers' aids

Starting from the history of English spelling, the chapter explains kinds of spelling errors and how to build a simple spell-checker (with Peter Norvig's spell-checker and dynamic programming for edit distance as Under the Hood topics). It introduces n-grams and simple generative language models, then grammar checking with phrase-structure and dependency grammar, descriptive versus prescriptive grammar, the formal complexity of languages, style checkers, and auto-complete and generative tools such as ChatGPT. The closing section links these ideas to the noisy channel model and later chapters.

Chapter 3: Computer-assisted language learning

This chapter compares first- and second-language learning and the factors behind L2 success, then asks why CALL systems need to be "intelligent". It discusses modelling language and modelling the learner, with two examples—FeedBook, an intelligent tutoring system built on a German seventh-grade English textbook, and Duolingo (including spaced repetition and A/B testing)—and ends with how such tools are evaluated and what they mean for language education.

Chapter 4: Text as data

The book pivots from knowledge-driven applications to letting data speak. It surveys text-as-data research in digital humanities, corpus linguistics, computational social science and authorship profiling, available English corpora and their metadata, and ethical issues such as privacy and bias. It then examines how words are distributed (function and content words, types and tokens, Zipf's power law and brevity law), topic modelling, and word meanings as vectors, from word-context matrices to contextualised representations; an Under the Hood box discusses the Voynich Manuscript.

Chapter 5: Text classification

Using spam filtering as the main example, the chapter explains how computers "learn" from labelled data, how to measure success (sensitivity, specificity, base rates and the base-rate fallacy, overfitting), and examples of classifiers including naive Bayes and the perceptron. It sketches transfer learning and the training tasks behind BERT, distinguishes supervised from unsupervised learning, treats sentiment analysis, and discusses benchmarks and leaderboards and their limitations.

Chapter 6: Searching

The chapter begins with an information need—finding an old friend online—and covers structured data, knowledge graphs and ontologies, question answering, Boolean search for professionals (with regular expressions and finite-state automata in Under the Hood boxes), and general web search: ranking by relevance (document vectors and TF-IDF) and by page quality, HTML, design choices and the future of search. It closes with recommendation systems and the social consequences of ranking, such as search engine optimisation.

Chapter 7: Machine translation

After examples of current machine translation quality and of translation needs, the chapter explores meaning across languages, typological differences and the translation triangle. It explains how systems are built from parallel corpora and word alignment, contrasts statistical and neural machine translation (including encoder–decoder attention and self-attention), introduces BLEU for evaluation, and reflects on why the task proved more tractable than once thought, recalling the 1966 ALPAC report.

Chapter 8: Dialog systems

Opening with HAL from 2001: A Space Odyssey, the chapter explains how conversations work—context, the Common Ground, non-declarative utterances, turn-taking, Grice's maxims, presupposition and implicature. It then covers chatbots (ELIZA, Microsoft's Tay, Kuki and ChatGPT, with the Chinese room thought experiment) and digital assistants (Siri, Google DialogFlow), including how each is built and evaluated, and the Turing test and Winograd schemas.

Chapter 9: Epilogue

The epilogue steps back to the economic, educational and socio-political consequences of language technology, framed as a debate between an imaginary pessimist and optimist, and poses questions for class discussion.

Key Concepts

Chapter 1: Encoding language on computers

• Language versus writing: writing is a technology for recording language, not language itself.
• Unicode and UTF-8: a universal inventory of characters and a scheme for encoding them as bytes.
• Spectrogram: a visual representation of the frequencies in speech over time.

Chapter 2: Writers' aids

• Edit distance: the minimum number of insertions, deletions and substitutions turning one string into another.
• n-gram language model: predicting the next word from the previous few, learned from data.
• Noisy channel model: inferring the intended message from a corrupted signal.

Chapter 3: Computer-assisted language learning

• Zone of proximal development: input pitched just beyond what a learner can already do.
• Learner modelling: tracking what a learner knows to adapt exercises and feedback.
• Spaced repetition: reviewing material at increasing intervals to aid retention.

Chapter 4: Text as data

• Zipf's laws: a few words are very frequent and most are rare; frequent words tend to be short.
• Topic modelling: discovering groups of co-occurring words that characterise documents.
• Word vectors: representing word meaning numerically by the contexts in which words occur.

Chapter 5: Text classification

• Sensitivity and specificity: how well a classifier finds the positives and avoids false positives.
• Base-rate fallacy: ignoring how rare a class is when interpreting a classifier's output.
• Transfer learning: reusing a model trained on one task (such as BERT's pre-training) for another.

Chapter 6: Searching

• Information need: what a user actually wants, which a query only approximates.
• TF-IDF: weighting a term by its frequency in a document and its rarity across documents.
• Knowledge graph: structured data linking entities and their relations.

Chapter 7: Machine translation

• Translation triangle: the idea that more abstract representations narrow the gap between languages.
• Word alignment: linking corresponding words in parallel sentences.
• Attention: the neural mechanism letting a translation model focus on relevant parts of the input.

Chapter 8: Dialog systems

• Common Ground: the shared information participants build up in conversation.
• Gricean maxims and implicature: the cooperative principles that let speakers convey more than they say.
• Winograd schema: a test sentence pair requiring world knowledge to resolve a pronoun.

Chapter 9: Epilogue

• Social consequences of language technology: effects on work, education and politics that the reader is asked to weigh from more than one side.

Critical Analysis

Strengths

• Truly introductory and inclusive. The preface states that the only background needed is curiosity about language and everyday computer use, and the book aims to welcome both "humanities-confident" and "computing-confident" students; technical depth is placed in optional "Under the Hood" boxes so the main text stays readable.
• Applications as the way in. Each chapter starts from a familiar tool—spell-checkers, Duolingo, spam filters, search engines, Google Translate, chatbots—and works back to the underlying linguistics and computation, showing how the same ideas (classification, n-grams, generate-and-rank, the noisy channel) recur.
• Up to date for an introductory text. The second edition covers word vectors, BERT, neural machine translation with attention, ChatGPT, benchmarks and leaderboards, so students meet current methods conceptually.
• Attention to ethics and consequences. Every chapter ends with a "Consequences" section, and topics such as corpus privacy and bias, search engine optimisation, Microsoft's Tay and the English focus of benchmarks are discussed alongside the technology; checklists and exercises support review.

Limitations

• No programming. The book explains how methods work but does not teach students to implement them; those who want hands-on skills need a separate programming-based text.
• Speech treated briefly. Speech recognition and synthesis are introduced in Chapter 1 and then set aside in favour of text.
• English-centred data. Although writing systems and typological differences are discussed, most worked examples and the corpora surveyed in Chapter 4 are English.
• Fast-moving topics. Sections on ChatGPT, digital assistants and search describe rapidly changing products and are likely to date quickly.

Where It Fits in the Course

In the Computational and Corpus Linguistics list, this is the conceptual entry point. It assumes no programming and explains what language technologies do and why they work, so it is a natural first read before Bird, Klein and Loper's Natural Language Processing with Python, which teaches students to implement similar analyses in Python. Doing Corpus Linguistics is generally described as a hands-on introduction to corpus methods and pairs well with Chapter 4 here.

McEnery and Hardie's Corpus Linguistics: Method, Theory and Practice and Gries's Quantitative Corpus Linguistics with R take the corpus-linguistic thread much further, methodologically and statistically. Jurafsky and Martin's Speech and Language Processing covers the algorithms behind Chapters 2 and 5–8 in technical depth; students who enjoy this book's "Under the Hood" boxes are ready to move towards it after gaining some programming and mathematics.

How to Use This Book

The book is written for undergraduates from any background, including linguistics, humanities and computer science students, and it grew out of university courses taught at Ohio State and Georgia Tech. Read Chapter 1 first, since it supplies the basic concepts of language, writing and encoding; Chapters 2–8 can then be taken largely in order, as later chapters build on n-grams (Chapter 2), vectors (Chapter 4) and classification (Chapter 5). The authors suggest reading Chapter 3 alongside Chapter 7 to compare human and machine multilingualism.

Use the margin keywords and the end-of-chapter checklists to review, choose exercises that interest you, and read the Under the Hood sections if you want more technical detail. For advanced students, the preface recommends combining the book with a reading list of research articles in digital humanities, computational social science and corpus linguistics for student presentations; the companion site (https://osf.io/v7uqm/) offers a sample syllabus and slides.

Edition and Sources

This guide is based on the second edition of Language and Computers (Language Science Press, 2024; Textbooks in Language Sciences 14; CC BY 4.0), read in the official open-access PDF linked from https://langsci-press.org/catalog/book/454 and hosted on Zenodo (DOI 10.5281/zenodo.12730906). It draws on the acknowledgments and preface (aims, audience, how to use the book), the full table of contents including every section and Under the Hood box, and for all nine chapters the opening sections, the "Consequences" sections and the end-of-chapter checklists.