Natural languages present a variety of challenges for parsing and information extraction, and for finding the crucial dependency relations that express the meaning of a sentence. Ideally, a parser should be able to adapt to a variety of different language types and constructions within a language, rather than being biased to one particular language, such as English. This book surveys in clear, precise terms the crucial relations in human languages, and how languages vary in the expression of these relations.

This short book consists of 100 concise chapters on natural language, focusing on the structure of words, morphology, and the building blocks of sentences, or syntax, as well as the interrelation of morphology and syntax. The chapters summarize what linguists know about the form of natural languages and how the formal features in a language map into the semantic composition of the meaning of sentences. This mapping is particularly relevant in natural language processing (NLP) for extracting information about the actors in an event, or “who did what to whom.”

The information in this book was first presented as a tutorial at the 2012 meeting of the Association for Computational Linguistics: Human Language Technologies section. The chapters follow a natural progression from general summaries to more detailed descriptions. The final chapter discusses existing resources for NLP. There is a full bibliography, with the page numbers where a given reference is cited. The index is detailed and clear, and there is a list of abbreviations and their full meaning. The text refers to a number of existing resources that are useful for NLP, such as WordNet, FrameNet, the Penn Treebank, and the World Atlas of Language Structures, a typological database.

The information about human languages that linguistics describes is often hard for the non-specialist to penetrate. What is important in language structures is often not immediately evident, and the technical vocabulary can be opaque. This book introduces these important concepts in a logical order, illustrated by familiar examples from English as well as other languages, ranging from relatively widely spoken languages like French, Mandarin, Turkish, and German, to quite exotic languages such as those of Native Americans and of Australia. One of the effects of the richness of example languages is that the singularity of English, a language with little inflection (much of it ambiguous) and fixed word order, is made clear. NLP systems often are constructed to be sensitive to the features of English, but for more general application they need to be able to read the features of languages of other types.

One of the goals of this book is typological, to illustrate the great variety found in human languages. Languages have elaborate prefixes and suffixes that indicate case and agreement with the subject and some or all of the objects, as well as free word order or systematical variations in the order of subject and object, as in active and passive voice, or subjects that are not expressed but understood. Some of the examples from Australian languages may strike the reader as structures that are unlikely to be encountered. But NLP programs may well need to deal with databases of examples of some of the Australian or Native American languages that are becoming extinct, in addition to large electronic corpora of major world languages.

The variations even in English are challenging for parsers, especially as there are interactions between constructions. For example, in (1a) below (a variation of an example on p. 111 in the book), “Kim” is the subject of “appears” in a complex sentence, but is the semantic patient of “plague,” as in (1b). In both (1a) and (1b), there are mismatches between the syntactic subjects and objects, and the semantic agents and patients, expressed more straightforwardly in (1c). Yet even this sentence has a semantically meaningless “it” subject.

(1a) Kim appears to be plagued by nightmares.

(1b) Nightmares appear to plague Kim.

(1c) It appears that nightmares plague Kim.

So even English is not free of challenges to parsing, due to systematic mismatches of form and function. These challenges from English and many other languages are explained in clear incremental stages, with copious cross-references.

The author is a member of the Department of Linguistics at the University of Washington, where she directs the MS program in computational linguistics. She is familiar with the analysis of natural languages from a variety of perspectives, including theoretical approaches specifically designed for computational application; however, the focus of this book is not on any one doctrine, but rather the whole range of information and previous work that could be helpful in constructing parsers. Her concern is to inform researchers of the aspects of a language that a parser might encounter in deriving structures and meanings, so that the parser is portable to a variety of languages.