Mach Translat (2014) 28:65–68DOI 10.1007/s10590-014-9149-9

BOOK REVIEW

Emily M. Bender: Linguistic fundamentals for naturallanguage processing: 100 essentials from morphologyand syntaxMorgan-Claypool, San Rafael, CA, USA, 2013, xviii + 166 pp

Francis Morton Tyers

Received: 12 November 2013 / Accepted: 18 February 2014 / Published online: 6 March 2014© Springer Science+Business Media Dordrecht 2014

Linguistic Fundamentals for Natural Language Processing is a compact referencework aimed at researchers and students in the field of NLP who do not have a back-ground in linguistics. As the subtitle, 100 Essentials from Morphology and Syntaxsuggests, the book mainly concerns itself with morphology and syntax. The author isEmily M. Bender, who is well known for her work on HPSG grammars. The bookhas a lot to offer readers of this journal, as many of the “fundamentals” she presentsare directly applicable to machine translation; what is more, in a few cases she pointsthe reader to recent MT research (e.g. on page 40 Le Nagard and Koehn 2010; and onpage 88 Shen et al. 2010) related to the problem at hand.

The book gives 100 fundamentals organised into ten chapters. Each fundamental isa short, one-paragraph to two-page explanation of a problem in linguistics. Chapter 1(10 pages) gives a general introduction to the book. Chapters 2–4 (42 pages) discussmorphology, while chapters 5–9 (70 pages) discuss syntax. Chapter 10 (4 pages)concludes the book by giving a short summary of linguistic and Natural languageprocessing (NLP) resources. The text as a whole has tried to be as suitable for beginnersas possible, however, particularly in the chapters on syntax, some linguistic terms areused without being defined, for example license when talking about verbs.

Each fundamental is generally illustrated with one or more examples from a widerange of languages. The examples are generally given in interlinear gloss format, andfor reference there is a glossary of abbreviations at the end of the book. I had two minorcomplaints with the examples. The first is that they contain a number of typographicalmistakes and printing errors, e.g. the Basque word lagunei ‘to friends’ is renderedlanguei (112). I counted a number of errors,1 but as I am not familiar with all of the

1 For a list of errata, see: http://xixona.dlsi.ua.es/~fran/100_fundamentals/errata.html.

F. M. Tyers (B)Alacant, Spaine-mail: ftyers@dlsi.ua.es

123

66 F. M. Tyers

languages in the examples, there may be more. My second complaint is that althoughthe author states that she will give examples first in “original orthography” (p. 2),there are a number of examples, e.g. those in Ingush, Chukchi, Kolyma Yukagir, andChechen, which are given only in transliteration.

1 Chapter 1: Introduction

The introduction contains seven fundamentals, which briefly describe morphologyand syntax, give an estimate as to how many languages there are in the world, andmotivate the value of linguistics to non-linguist NLP practitioners. The fundamentalon the number of languages goes into the problems of making such an estimate. Theauthor also points out, quite rightly, that as the vast majority of work is done onvery few languages, there can be doubts as to the “language independence” of certainlinguistically-uninformed approaches to NLP, and invites readers to consider if theNLP systems that they are working on would function as well for all of the languagesin the examples in the book.

2 Chapters 2–4: Morphology

There are three chapters on morphology: Chapter 2: Introduction, Chapter 3: Mor-phophonology and Chapter 4: Morphosyntax. The introduction to morphology givesfifteen fundamentals, providing a succint overview of the field of morphology, start-ing from “what is a morpheme” and continuing to describe inflection and derivation,morphotactics—constraints on morpheme order—and different morpheme and mor-phological types. The main thing that is missing from this chapter is a descriptionof intra-language variation in morphology and orthography (e.g. -ise, -ize; shown,showed) and normative versus non-normative morphology and orthography (e.g.aren’t, ain’t; because, ‘cos). The short, five-page, chapter on morphophonology givesa very brief overview of the field, describing the relationship of underlying forms(e.g. strings of morphemes) to pronunciation and orthography. All in all, as the authorremarks, for NLP practitioners working mostly with text, which is the case for MT, thediscussion of morphophonology will only be relevant to the extent that it is representedin the orthography. Given this it is surprising that she does not mention the fact thata single language may have several orthographies, which may differ in terms of howthey represent the morphophonology.

The final chapter of the first section describes morphosyntax. This is in my opinionthe most lucid chapter of the whole book. The fundamentals presented cover manylinguistic properties that can be represented morphologically: tense, aspect, mood;person, gender, number; case; negation; evidentiality; definiteness; honorifics; andpossession. This discussion is followed by a number of fundamentals on agreementof these properties and differences between languages. In the fundamental discussinggender, number and case, the work by Le Nagard and Koehn (2010) on improvingmachine translation of pronouns using co-reference resolution is mentioned.

123

Emily M Bender, Linguistic Fundamentals 67

3 Chapters 5–9: Syntax

The five chapters on syntax are: Chapter 5: Introduction; Chapter 6: Parts of speech;Chapter 7: Heads, arguments and adjuncts; Chapter 8: Argument types and grammat-ical functions; and Chapter 9: Mismatches between syntactic position and semanticroles.

The introduction is short (three pages), and basically describes syntax as puttingconstraints on possible sentences, and states that it is the “scaffolding” for semanticcomposition, something that is relevant for various approaches to machine translation.Chapter 6 gives a very brief four-vignette introduction to parts of speech, of which twovignettes describe how parts of speech can be defined distributionally or functionally.The arguments are convincing, but it is unfortunate that there is no description or men-tion of part-of-speech tagging—neither rule-based nor statistical—nor how linguisticfundamentals could be of value in improving resources for training part-of-speechtaggers (see Manning 2011).

While the previous chapters were directly applicable to NLP systems, presentingfundamentals and examples which might have an effect on NLP systems, the threemain chapters on syntax (7, 8 and 9) seem to only present theoretical syntax in general.A major problem with these three chapters is that the examples are often contrived andunrealistic, including some (I counted five) examples that I would consider grammat-ical but marked as ungrammatical or vice versa. For example in (86), “People rumourthem to be CIA agents” is marked as ungrammatical in that “rumour” only has a pas-sive form; however, a search in a popular search engine shows thousands of activeuses of “rumour”, and also does not sound ungrammatical to me as a native speakerof English. This sort of problem could have been avoided by selecting real-worldexamples (e.g. from a corpus) and testing with native speakers. A further quibble isthat the vast majority of examples are in English—105 out of the 136 in this section.MT practitioners might wonder how frequent some of the phenomena are, and howimportant they are for building effective machine translation systems, either corpus-based or rule-based. A further issue is that these chapters can be somewhat crypticfor the non-syntactician. For example, a lot of terminology is used without definition.At the very least, a glossary of such terms could have been useful, though a thoroughtreatment of how these terms should be understood would also have been beneficialfor the audience the book is aimed at.

Chapter 7 focusses on defining heads, arguments and adjuncts and their behaviour,describing what constituency structure is and how arguments can be distinguished fromadjuncts. That groups of words form constituents is presented as fact, and dissentingviews, such as Evans and Levinson (2009, §5), who question the primary of hierarchicalmodels, are not discussed. The descriptions are clear, although the vignette describingthe difference between an argument and an adjunct could benefit from an example.Also, all but 3 of the examples are in English. This is followed up by Chapter 8,which deals with argument types and grammatical (syntactic) functions. It starts bydescribing different semantic roles and syntactic functions and relating them to eachother. It continues by giving a description of how word order, agreement and case canbe used to identify syntactic function and how morphosyntax can change syntacticfunction. In discussing semantic dependencies, the work by Shen et al. (2010) on

123

68 F. M. Tyers

string to dependency statistical MT is cited. The final chapter in this section, Chapter9, describes mismatches between syntactic function and semantic role, for examplein passives and impersonal constructions. It also contains fundamentals describingcoordination, long-distance dependencies and argument drop.

4 Chapter 10: Resources

The final chapter (98–100) gives an overview of some linguistic resources which couldbe useful to NLP researchers. It describes morphological analysers, syntactic parsersand a typological database. The description of morphological analysers is adequate,but the author strangely only gives two examples for Japanese and Arabic. It wouldhave been good to include a few more, or at least give pointers to online directories.At the other end of the analysis chain, the author describes “deep” semantic parsersin the HPSG and LFG frameworks. There is no mention of part-of-speech taggers,nor shallow-parsers, nor less “deep” syntactic parsers—where there exist numerousfree/open-source resources and resource pools for various languages (Streiter et al.2006).

5 Summary

In general Linguistic Fundamentals for Natural Language Processing is a good ref-erence text for linguistics. The layout is very convenient for quick reference. Whileother introductions to linguistics may be aimed specifically at students of linguisticsor a general audience—for example Larry Trask’s Introducing Linguistics (Trask andMayblin 2005)—this work is targetted specifically at researchers in NLP, particularlythose from a non-linguistics background. The book generally succeeds in its aims andhas a lot to offer. However, I certainly think that there is room for a similar work, moredirected at MT, and with a more diverse coverage of languages, and more practical,corpus-based, coverage of syntax.

References

Manning CD (2011) Part-of-speech tagging from 97% to 100%: is it time for some linguistics? In: GelbukhA (ed) Computational linguistics and intelligent text processing, 12th international conference, CICLing2011, proceedings, part I. Lecture notes in computer science, vol 6608, pp 171–189

Evans N, Levinson S (2009) The myth of language universals: language diversity and its importance forcognitive science. Behav Brain Sci 32:429–492

Le Nagard R, Koehn P (2010) Aiding pronoun translation with co-reference resolution. In: Proceedings ofthe 5th Workshop on Statistical Machine Translation, pp 252–261

Shen L, Xu J, Weischedel R (2010) String-to-dependency statistical machine translation. Comput Linguistics36(4):649–671

Streiter O, Scannell K, Stuflesser M (2006) Implementing NLP projects for noncentral languages: instruc-tions for funding bodies, strategies for developers. Mach Transl 20(4):267–289

Trask L, Mayblin B (2005) Introducing linguistics: a graphic guide. Icon books, UK

123