Zero affixes in derivational morphology: Introduction

1.1 The empirical challenge

Natural language knowingly makes rich use of morphologically identical items for different functions. Inflectionally poor languages like English often show paradigmatic syncretisms such as in (1), where the present tense is overtly realized only in the third person singular form. But even when the language has a morpheme dedicated to a particular function such as the nominal plural in (2a), this may fail to be overtly realized in some items as in (2b):

If the absence of a signifier challenges the traditional conception of the Saussurean sign, the postulation of zero morphemes for the corresponding functions, as illustrated in (1) and (2), facilitates a way to maintain the one-to-one mapping between form and meaning in puzzling cases in which a difference in meaning is not accompanied by a phonological form.

Besides inflectional paradigms, zero affixes have largely been postulated for derivational processes, which represent the focus of this collection. The standard phenomenon here is that of category change, illustrated in (3), but zero morphology has also been employed to account for labile verbs in argument structure alternations such as the causative-inchoative illustrated in (4). More complex constructions such as the Italian nominalized infinitive in (5) may also be argued to involve zero affixation at the interface between morphology and syntax.

While in languages like English, the two forms related by a zero affix are perfectly identical as in (1)–(4), inflectionally richer languages like Italian or German may show small formal differences triggered by phonological changes such as ablaut and umlaut in (6) and by declension classes with gender marking as in (7).

Besides such formal differences, the productivity of morphological phenomena that involve zero affixes, in general, is higher in languages like English than in inflectionally richer languages. The reasons for this difference relate to the morphological type of the language, language-specific morphophonological constraints and the status of the base (root vs. stem vs. word) (Bauer and Valera 2005; Don 2005; Valera 2015).

1.2 The theoretical debate

While the advantage of positing zero affixes is straightforward from an empirical perspective, their theoretical legitimacy has been controversially debated for several decades already (see Lieber 1981; 1992; 2004; Kiparsky 1982; Anderson 1992; Don 1993; 2005; Pesetsky 1995; Plag 1999; Mel’čuk 2002; Borer 2013 and the recent overview in Dahl and Fábregas 2018). In general, positing a contentful zero morpheme confronts us with the theoretical quandary of differentiating it from the lack of a morpheme altogether: how do we know where there is a zero and where there is nothing?

A reasonable condition for positing a zero affix is the existence of an overt affix with a similar function (i. e. its “overt analogue” in Sanders 1988). In inflectional paradigms one easily finds such evidence, as shown in (1), but for derivation it is not always straightforward. For conversion from verbs to nouns in English, for instance, we find zero and overtly marked pairs such as in (8), but it is debatable whether this zero affix has a coherent enough meaning to be retrieved from all such deverbal nouns: the noun climb-Ø denotes an event or a path, break-Ø may describe an event, a state or a result entity, and cook-Ø unambiguously denotes an agent, an interpretation otherwise contributed by the participant-denoting -er suffix. Therefore, Ø emerges here as a nominalizing suffix with a very polysemous meaning.

Following a strong interpretation of the “overt analogue criterion” (Sanders 1988) for English denominal verbs as in (3a), it has been argued, for instance, that a zero suffix would be semantically too diverse to form a lexical entry by comparison with overt verbalizing suffixes such as -ify and -ize (Plag 1999; Lieber 2004). Another reservation about allowing zero affixes comes from the undesired effect of zero proliferation and the indeterminability of their ordering in the absence of sound criteria for their existence or interaction with each other (Anderson 1992; Bergenholtz and Mugdan 2000; Dahl and Fábregas 2018). However, in separationist theories such as Distributed Morphology (Halle and Marantz 1993; Marantz 2013; cf. Beard 1995), zero affixes represent a morphophonological reality, which is distinct from the morphosyntax and the interpretation of complex words. No lexical entries need to be posited for them, and their ordering comes from the underlying morphosyntax and semantics.

Among the various mechanisms that have been put forward as an alternative to zero affixation, we find relisting, underspecification and cumulative exponence. Relisting is a form of lexical coinage triggered by pragmatic needs and has been proposed by Lieber (1981; 1992; 2004) for conversion phenomena. In underspecification theories, the lexical category of words in conversion pairs gets specified only in a syntactic context (see Farrell 2001 and Borer 2013). In her syntax-based Exo-Skeletal Model, for instance, Borer (2013) takes both zero-derived verbs and nouns as in (3a, b) to be formed by means of a verbal/nominal external projection such as T(ense) and D(eterminer) attached to a category neutral root. Whether it is a verb or a noun that ensues depends strictly on the categorial nature of this syntactic projection.

Zero morphemes are also avoided in frameworks like A-Morphous Morphology (Anderson 1992) or Word-and-Paradigm models (Blevins 2016; Bonami and Strnadová 2019), which dispense with traditional morphemes altogether. In these theories, cumulative exponence is argued for cases where a word form expresses more than one morphosyntactic feature but not all features have a corresponding (sublexical) exponent. Zero exponence receives yet another explanation in frameworks such as Brody’s Mirror Theory (2000) or Starke’s Nanosyntax (2009), allowing a single morphophonological exponent to lexicalize multiple morphosyntactic features. Specifically, in Nanosyntax mechanisms like phrasal spellout (i. e. the insertion of vocabulary items into nonterminal nodes) and spanning (i. e. the lexicalization of a sequence of heads by a single vocabulary item) are deployed to explain the cumulative exponence of portmanteau morphemes, avoiding the postulation of zero morphemes (Caha 2018; Taraldsen 2018).

However, it has been argued that such mechanisms fail to capture relevant morphosyntactic and phonological constraints which would be explained under a zero affix approach (see Don 1993; 2005; Darby 2015). For instance, in English denominal zero-derived verbs as in (3a) are fully productive, while deverbal zero-derived nouns as in (3b) are more limited. Treating both by means of the same mechanism does not account for this difference, while two different zero affixes would account for the different productivity levels.

Directionality has often been invoked in support of a zero affix derivational approach for category change phenomena, and, depending on it, proposals have been made that some formations may require a zero affix analysis, while others should be accounted for by other means. Lieber (1981: ch. 3) first proposed to treat typical category change as a non-directional process and to employ zero affixes only exceptionally for morphologically systematic formations such as adjectival participles built from verbal participles. More recently, Darby (2015) and Darby and Lahiri (2016) have used psycholinguistic evidence to argue that in category change some noun-verb pairs show clear directional behavior, while others are underspecified. This suggests the need for two approaches: a derivational and an underspecified approach, respectively. Further support for the need of both approaches is also provided by the computational study in Kisselew et al. (2016). From their investigation of distributional semantic models, they conclude that zero-derived nouns in English lend themselves to a derivational analysis, while the behavior of zero-derived verbs conforms to an underspecified analysis.

Experimental support for derivational zero affixes in general has also been provided by neurolinguistic studies. Pliatsikas et al. (2014) reports the results of an fMRI experiment where they compared the processing of one-step (soaking < soak-V) and two-step (bridging < bridge-V < bridge-N) morphologically derived words. Interestingly, they found that, despite similar surface structure, two-step derivations trigger higher processing costs than one-step derivations, hence providing critical evidence for the morphological complexity of zero-derived forms as processed in the brain. These results are also confirmed by the psycholinguistic studies of Darby (2015) and Darby and Lahiri (2016), which employ lexical decision tasks with masked and delayed priming.