This article reexamines the Brown Corpus as a landmark resource in corpus linguistics and a practical basis for comparative research and data-driven language education. Drawing on the official corpus manual, publications, and the Brown family of comparable corpora, it reviews the corpus's compilation principles, sampling frame, genre composition, versions, annotation, and technological production. It also synthesizes five representative studies of sentence complexity, modal change, genitive alternation, relative clause variation, and word-frequency norms. The review shows that the corpus's enduring value lies less in its one-million-word size than in its transparent sampling design and replicability across regional and diachronic corpora, including LOB, Frown, FLOB, and AmE06. At the same time, its 1961 publication date, restriction to edited prose, unequal genre sizes, excerpt-based structure, and version-dependent tokenization limit generalization and require documentation of analytical conditions. Three AntConc activities are consequently proposed to connect corpus design with hands-on inquiry: analyzing the meanings and genre distribution of must, comparing American and British English through Brown and LOB keywords, and distinguishing the near-synonyms begin and start. The article argues that Brown remains valuable when treated not as a model of present-day English, but as a controlled historical benchmark for reproducible comparison and corpus literacy.

This article reexamines the British Academic Spoken English Corpus (BASE) as a resource for research on university lectures and seminars and corpus-based language education. Drawing on the BASE manual, corpus documentation, and five representative studies, it reviews BASE's compilation, disciplinary balance, transcription conventions, versions, research applications, and pedagogical potential. The official release contains 1,644,942 tokens from 160 lectures and 39 seminars recorded at two British universities between 1998 and 2005. The review shows that BASE supports research on lexical bundles, pragmatic functions of just, relevance markers, lexical coverage, and tag questions, linking vocabulary, discourse organization, and lecturer-student interaction. Its strengths include coverage of four disciplinary domains, inclusion of both lectures and seminars, and annotation of speaker and nonverbal information. However, differences among the original XML files, research-specific datasets, and platform versions affect document and token counts. Uneven annotation, limited seminar data, normalized transcription, and restriction to two institutions constrain interpretation. Three AntConc activities are proposed: comparing just across lectures and seminars, contrasting BASE with MICASE, and analyzing relevance markers. The article argues that BASE remains valuable when researchers document versions and preprocessing decisions and treat findings as evidence about particular British academic settings rather than academic speech in general.

This article reexamines the British Academic Written English Corpus (BAWE) as a resource for investigating university assignments and supporting corpus-based writing education. Drawing on corpus documentation and five studies, it reviews BAWE's compilation, stratified sampling, analytical units, genre classification, metadata, versions, research applications, and pedagogical potential. BAWE contains 2,761 assignments and 6,506,995 words produced in the early 2000s by undergraduate and taught master's students at several British universities. The review shows that university writing extends well beyond essays to thirteen genre families and that disciplinary and genre differences may outweigh level-related variation. Research using BAWE indicates that advanced writing may rely more on complex noun phrases than subordinate clauses, that lexical bundles reveal disciplinary relationships, and that successful writers with different first-language backgrounds use cohesive resources in varied ways. Strengths include authentic assessed writing and rich contextual metadata. However, differences among assignments, independent texts, genre units, and corpus versions complicate replication. Selection of higher-rated work, uneven subgroup sizes, and limited temporal and institutional coverage constrain generalization. Three AntConc activities are proposed: comparing first-person expressions across genres, contrasting BAWE with MICUSP, and developing discipline-specific writing guides. BAWE remains valuable when findings are interpreted within genre, discipline, level, and assessment contexts.

TagAnt is a freeware multilingual Part-Of-Speech (POS) tagging and lemmatization tool developed by Laurence Anthony. By integrating the SpaCy NLP framework, it supports over 25 languages and merges text segmentation features into a unified interface. It offers intuitive language model management, direct processing of TXT, DOCX, and PDF files, and standard tagsets like Penn Treebank. While limitations exist, such as memory overload with large texts in the UI and domain-specific tagging errors in L2 learner corpora, batch processing and a user-friendly GUI mitigate these issues. TagAnt significantly lowers the barrier to corpus linguistics for researchers and educators worldwide.

This review examines Machine Learning in Translation, edited by Peng Wang and David B. Sawyer, as an accessible, human-centered introduction to the integration of machine learning into translation theory, practice, and education. Organized into three parts, the book connects human and machine approaches to translation, explains key technologies and tasks, including neural machine translation, quality assessment, and natural language processing, and highlights the central role of language data in translator-computer interaction. Particular attention is given to translators' intentionality, model customization and personalization, machine learning literacy, and evolving professional competences. The book's main strength lies in presenting machine learning as a continuation of linguistic, communicative, and cognitive approaches rather than merely an engineering domain. However, its predominantly conceptual orientation, limited empirical evidence, emphasis on resource-rich languages, and brief treatment of ethics and implementation restrict its scope. Overall, the volume provides a valuable roadmap for translation practitioners, educators, and researchers navigating increasingly data-driven, human-machine translation environments and future interdisciplinary collaboration.