Computational linguistics turns the descriptive stack (morphology, syntax, semantics, pragmatics) into something that can be applied automatically, enriching raw text with linguistic information that downstream social-science analysis can repurpose. It is an interdisciplinary field that combines linguistics, computer science, artificial intelligence, and mathematics to process and model natural language. Its first methodological idea is distant reading. Instead of reading texts (closely), one applies quantitative procedures to large collections and looks at patterns. The material for such work is the corpus, a systematic, structured, and machine-readable collection of texts or transcribed speech, and the questions it can answer, from frequency and concordance to language change, depend on how the corpus was built.
Counting words, literally, shows why linguistic theory matters for computation. Before anything can be counted, one must decide what counts as a word, and the answer depends on the level of analysis. Is Haustür one word or two, are gehen, ging and gegangen one item or three, and are Der and der the same? Each choice corresponds to a different unit, such as the token, the type, or the lemma, and changes the result. The ratio of types to tokens measures lexical variety, and frequency profiles can support stylometry and authorship attribution.
A word cloud makes the point visually. Raw counts are dominated by stopwords and punctuation, which carry little content, so analysts filter them or restrict the analysis to certain word classes, such as verbs for actions, adjectives for evaluation and nouns for topics. Every filter is an implicit theory of what matters in a text. Named-entity recognition goes a step further and detects real-world objects such as persons, organisations, places and dates, which links the text to entities outside language, a move that echoes the referential ideas of semantics.
The work of a computational linguist can itself be described as a language-transformation pipeline, essentially a sequence of decisions about abstraction. Raw text is collected, cleaned, pre-processed, turned into numerical features, modelled, evaluated, deployed and monitored. At each step, some information is discarded, and some structure is imposed, so that a computational model’s behavior/performance is determined by the aforementioned steps.