1.1 Zero Morphology: A First Approach

Consider an English irregular plural like (2).

(2)

The example in (2) involves zero morphemes for some theories, while other theories will not posit such objects (cf. section 3 for the causes and consequences of this choice in specific theories). (3) is one example of an analysis involving zero morphemes (cf. Halle & Marantz, 1993; Pfau, 2000; and Embick & Halle, 2005 for analyses along these lines).

(3)

The core idea in (3) is that the singular noun mouse, which by hypothesis underlies the plural in (2), alters its shape through a readjustment rule triggered in the context of the plural morpheme, which is a zero morpheme devoid of phonological content. Alternatively, it could be proposed that (2) is a root that is only licensed under [plural] (à la Harley & Noyer, 2000; cf. 4), which makes the readjustment rule unnecessary.

(4)

This is not the only option when analyzing (2). An existing alternative is to propose that mice is a cumulative exponent—otherwise known as a ‘“synthetic form’” or ‘“portmanteau morpheme’”—that spells out collectively the nominal features and the plural features (5).

(5)

This use of cumulative exponence is made in all theories where words are not decomposed into morphemes (Item-and-Process and Word-and-Paradigm theories; see in particular Anderson, 1992, p. 61 for an influential critique of the use of zero morphemes), but also in some that posit morphemes as the building blocks of words, but do not assume that each exponent must necessarily correspond to a single morphosyntactic head. Siddiqi (2006, pp. 56–59) adopts a proposal along these lines using the terms of Fusion in Distributed Morphology’ (see Bobaljik, 2017; cf. also Noyer, 1997). Relevant here are also other procedures that allow through some technical means that a single morphophonological exponent spells out more than one morphosyntactic unit: cf. Brody’s (2000) Mirror Theory; also, Ramchand’s (2008), Dékány’s (2012), and Svenonius’ (2016) notion of Spanning, and finally Starke’s (2009), Caha’s (2009), and Markus’ (2015) notion of Phrasal Spell Out. It is more controversial whether Hale and Keyser’s (1993, 2002) notion of Conflation involves zero morphemes or not, given the historical evolution in how the concept has been defined by the authors themselves.

1.2 Zero Morphology Versus Null Phonological Marking

The comparison between (3)–(4) and (5) shows that zero morphology is distinct not only from cumulative exponence in the wide sense, but also from zero phonological marking. The analytical procedure in (3) treats (2) as involving zero morphemes because it proposes that the phonological change that the base exhibits is not directly codified in the plural morpheme, but is the result of a readjustment rule. An alternative analysis could have proposed that the plural morpheme is not phonologically empty, but on its phonological side comes with floating features responsible for the change in the vowel quality of the base.

(6)

Analyses that treat irregular morphology in this way have been criticized (cf. Bermúdez-Otero, 2012), but they are mentioned in this context to highlight that whether a structure involves zero morphology or not depends on an intricate network of analytical choices; section 3 is devoted to discussing what these choices are.

In the same way that the presence of phonological marking does not exclude an analysis involving zero morphology in absolute terms, the surface absence of phonological marking does not force such a type of analysis. Consider (7), from Spanish.

(7)

The word corresponding to Monday is identical in the singular and in the plural, even though (7b) acts as morphosyntactically plural for agreement processes (cf. the determiner and copular verb forms). Perhaps some analyses would be inclined to propose a decomposition of (7b) along the lines of (8), that is, involving a zero morpheme for [plural].

(8)

However, this is not forced even in this case. The unmarked allomorph for plural in Spanish is /s/. Assume the noun lunes takes this allomorph, as in (9). Spanish lacks geminate consonants, so */ss/ is a phonologically impossible sequence in the language. The simplification to /s/ makes the plural form surface identical to the singular, but note that no zero morpheme was used here.

(9)

Is there an empirical way to decide between (8) and (9)? In this case, an argument can be made in favor of (9): all words ending in a stressless syllable whose final segment is /s/ follow the pattern of lunes: tesis ‘thesis,’ análisis ‘analysis,’ trilobites ‘trilobite,’ etc. The analysis in (8) would treat it as a coincidence that the nouns that select the zero morpheme for plural form a natural class from the perspective of their phonological shape.

1.3 Zero Morphology Versus Null Morphological Representation

The observations made in the previous section extend beyond phonological marking; zero morphology should also be differentiated from absence of morphological representation.

In morphological research, it is accepted that distinct languages might grammaticalize different properties in the form of features. Gender systems and agreement (Corbett, 1991, 2006) are a prime example of this. Similarly, Chinese lacks surface morphemes to mark agreement, but it is standardly assumed that Chinese belongs to a type of language where agreement features are not grammaticalized, not that this language extensively uses zero morphology.

When faced with a situation where marking is lacking from the surface, it is necessary to have criteria to determine whether marking is lacking because the property has not been grammaticalized in the language, or because it is grammaticalized but the morphology is null—for the sake of the argument, assuming a system where zero morphemes are allowed. In some cases, a principle of systematicity or homogeneity could apply: if in all forms of a verbal paradigm except for one the marking for second person singular is overt, it is in principle plausible to think that the property has been grammaticalized in all forms, and consequently that the unmarked form in actuality carries a zero morpheme.

The plausibility is not the same, however, if the form that lacks the second person singular marking has chances to be structurally different from the other forms, as in the following case of the Portuguese imperative:

(10)

Zero morphemes are still a possibility here, but given the unique syntactic properties of imperatives, the option that person agreement is not grammaticalized just in (10d) is not completely implausible. English, a non-pro-drop language, happens to allow pro-drop precisely in the imperative (cf. the gloss in 10d), which has suggested to some authors that the structure of imperatives lacks the layers responsible for introducing subjects, and therefore also subject agreement (cf. Zanuttini, 1997; Biezma, 2009; the general observation is that imperative forms tend to be associated with stem of root forms cross-linguistically). Thus, it might be the case that person agreement is not expressed precisely in one form of the verb, and thus zero morphemes could be avoided here as well.

The conclusion is that independent, construction-specific tests are required to determine whether a property is grammaticalized or not. One example of this type of research is the debate on whether Jamaican Creole grammaticalizes passive. As (11) shows, on the surface there is no morphological difference between active and passive.

(11)

Part of the literature on this language has argued that Jamaican Creole simply does not grammaticalize passive (e.g., Roberts, 1993). Lacharité and Wellington (1999), in contrast, argue that passive is expressed in the language, only with zero marking. Their evidence is that in (11b), as is the case with standardly marked English-style passives, the subject is marked as a patient (Jamaican Creole is not pro-drop), the structure does not accept a direct object, and an anticausative interpretation is not allowed. Syntactically and semantically, there are reasons to think that Jamaican Creole distinguishes passive from active, so the zero-morpheme hypothesis is more plausible in this case—again, assuming a theory that allows such objects.

The delimitation problem can be even more nuanced if we take into account a number of theory-internal distinctions that are central in Trommer’s (2012) discussion of zero-exponence. Trommer points out that beyond the possibility that some property is not grammaticalized through a morpheme, some theories allow for the option that a property is grammaticalized but does not receive morphological expression. Some theories like Distributed Morphology (Halle & Marantz, 1993) and Optimality Theory (Prince & Smolensky, 2004 [1993]) allow that the mapping between syntax and morphology is not trivial. Specifically, Distributed Morphology allows for the possibility that the features that in principle would be expressed by a morpheme are erased before a particular exponent can be introduced. The operation is called impoverishment (Bonet, 1991) or obliteration (Arregi & Nevins, 2012), depending on whether what is erased is the feature-content of a terminal node or the terminal node itself. Within Optimality Theory, a violable constraint Realize Morpheme (Kurisu, 2001) has been proposed, which again opens up the possibility that a property that is grammaticalized in the language is not expressed morphologically.

The difference between this situation and zero morphology is theoretically subtle. In the case of zero exponence, a morpheme is used to express a distinction, but its phonological shape is zero. In the case of morphological non-expression, a distinction is grammaticalized but not linked to any morpheme. If the distinction is subtle, so are the criteria used within theories to decide between the two. Morphological non-expression is preferred as a solution in Distributed Morphology when there are conceivable morphological principles that can be violated by the presence of the missing morpheme. Take Pescarini (2011), who analyzes the ungrammaticality of (12a) in Italian. (12a) exhibits two morphologically identical clitics ci that are syntactically distinct: one is a 1pl object clitic ‘us,’ while the other is a locative clitic ‘there.’ The sequence of two adjacent identical clitics is solved by having only one ci, something that Pescarini (2011) relates to the presence of a morphological filter that bans sequences of identical clitics (a form of generalized haplology, cf. Nevins, 2012; see also Perlmutter, 1971 for Spanish and Bonet, 1991 for Catalan and Spanish). Theory-internally, the problem is not phonological, and therefore the proposed solution within Distributed Morphology is morphological non-expression rather than zero morphology.

(12)

Table 1 is meant to present in a schematic fashion the four senses of absence of marking that we have discussed in this section. Only the one that involves morphological expression by an item deprived of a phonological representation counts as zero morphology in the strict sense.

2.1 Theories That Do Not Allow Zero Morphemes

What assumptions have to be made in order to block the existence of zero morphemes from the theory in its entirety? In the narrow sense that we have defined above, zero morphology involves a situation where we have a linguistic unit with content but with a null form. By the commonly assumed principle of arbitrariness of the sign, which states that the meaning of a unit does not directly condition its phonological shape, such a situation should be allowed: there are no direct correlations about how content must map to sound, so it is theoretically possible that some content is linked to an empty phonological representation.

Unsurprisingly, then, a theory that recognizes morphemes as units but blocks the possibility that their marking is null must a be a semiotically oriented theory which assumes a principle of iconicity that mediates between content and form. One example of such a theory is the specific version of iconicity adopted in Natural Morphology (Dressler, Mayerthaler, Panagl, & Wurzel, 1987). Constructional Iconicity or Diagrammaticity (Wildgen, 1982; Dressler, 1987) states that an increase in the content associated to a word must be accompanied by a proportional increase in the form of that word (see also Wunderlich & Fabri, 1995 for a related critique of zero morphology). This iconicity principle has been considered a violable natural tendency that in fact has a few counterexamples: see (14) (Mayerthaler, 1987, p. 49), and also Manova and Dressler (2005) on how language-internal principles might supersede iconicity. However, assuming some sort of iconicity principle as part of the theory is necessary to set a system where zero exponence is theoretically impossible.

(14)

However, such a principle is not enough to make zero morphology unavailable in all cases. There are several cases in which, empirically, a particular value of a feature is not phonologically signaled in the word, which displays an unmarked shape. Mayerthaler (1987, p. 48) and Bybee (1991, p. 71) note that among the typically unmarked values cross-linguistically one finds singular number and nominative case for nouns, positive degree for adjectives, and present, active, and indicative for verbs. How does one avoid zero morphology in such cases? The additional assumption that one has to make in such cases is that these values are the default lexical interpretations of the grammatical categories; that is, that without further information, nouns are interpreted as singular nominatives, verbs as present indicative, and adjectives as being in the positive degree. The values that are unmarked are argued to be cognitively favored in some way, so they can be assumed in the absence of contradicting information; to give a concrete example, if verbs are used to present information about eventualities in the world, this hypothesis assumes that talking about eventualities that are actual in the real world (indicative) is cognitively unmarked with respect to hypothetical situations (subjunctive). In this way, instead of a representation like (15a), which would involve a zero exponent that violates diagrammatic iconicity, we would have a representation like (15b), complemented with an interpretative instruction stating that verbs are by default interpreted in particular values corresponding to “natural” cognitive categories.

(15)

However, it is fair to say that these two assumptions that are central in Natural Morphology have not been assumed by other theoretical frameworks. If a theory does not enforce an inviolable Diagrammatic Iconicity Principle and does not assume that some morphosyntactic values will not receive marking because they are cognitively assumed by default, zero morphology is in fact allowed.

2.2 Theories That Allow Zero Morphemes, but Discourage Them on Methodological Grounds

The vast majority of existing theories adhere to a form of the arbitrariness of the sign, and therefore to a non-deterministic mapping between meaning and form, and as such they open the door for the existence of zero morphology. However, that something is possible does not imply that its existence is accepted without discussion, and in fact many proposals still condemn the use of zero morphemes in analyses.

There are two ways in which zero morphemes can be condemned in theories that accept a form of arbitrariness of the sign. The first one is a methodological concern: the claim is that any analysis that uses zero morphemes opens the door for unfalsifiable claims. Anderson (1992, p. 71) illustrated two of those problems with his discussion of (16): proliferation of zero forms and indeterminability of ordering.

(16)

According to Anderson, in (16), an analysis that allows zero morphemes risks positing five such morphemes: one to mark the conjugation class, another one for indicative mood, a third for active voice, a fourth for present tense, and a fifth for first person or singular number, depending on how the subject agreement -o is analyzed (cf. also Scalise’s 1984 remarks on participial forms within zero morpheme analyses). This proliferation has been argued to be an overstatement by Halle and Marantz (1993, p. 133): it is not necessarily the case that there is one separate morpheme for each feature, and in fact the Latin conjugation tends to bundle some of these properties together (for instance, person and number: -s is associated to 2nd singular, -mus to 1st plural, etc.).

But there is a second general problem also pointed out by Anderson: if there are several zero morphemes in (16), what empirical criterion could be used to determine how they are ordered with respect to one another? In principle, given the lack of a phonological representation, any of the orderings in (17) could in fact be proposed; with seven morphemes, two overt and five null, that makes 5,040 possible orderings.

(17)

Of course, there are conceivable arguments to choose among these orderings: when the morphemes are overt, Latin does not mark any of these properties with a prefix, and the ordering between the overt morphemes can also help decide between the options both with respect to their linear placement and which properties tend to be bundled inside the same morpheme. But that is not Anderson’s point; the point is that once zero morphemes are allowed into the system, any claim about their number, placement, and internal organization is not motivated by direct empirical evidence but by analogy with the overt forms, sometimes on the assumption that there is a high degree of systematicity within a paradigm.

These methodological concerns are common in the literature; see, as an illustration among many that make the same point, Stewart and Stump (2007, p. 389):

While the use of phonetically null affixes is not new, it is nevertheless a questionable formal device, if only because the putative distribution of such affixes is hard to demonstrate empirically. Often, zero affixes arise in a Structuralist implication on analogy with the distribution of one or more overt affixes with comparable but contrastive meaning. [...] Taken to its logical conclusion, this move engenders either a large population of homophonous null affixes or a potentially long derivation of string-vacuous rule applications. In either case, the argument is developed theory-internally, and it is therefore unfalsifiable.

The second way in which theories that otherwise allow a mismatch between meaning and form ban zero morphemes is through performance-based or acquisition-based concerns. The general idea is that zero morphemes, while theoretically possible, should not be allowed—or at least should be extremely minimized—because they cause trouble for processing, acquisition, or memory load. Consider, for instance, Siddiqi (2006, pp. 70–71), who notes that in the framework he assumes, Distributed Morphology, zero morphemes have to be used to a greater extent than in other theories, in part due to the assumption that roots lack a grammatical category and they must have one assigned by a functional head dominating them. He proposes a technical device along the lines of example (5) to reduce the number of zeroes, and points out that his approach is superior to one where zero morphemes proliferate because it reduces both the computational load and the memory load. The memory load is increased by the presence of zero morphemes because a zero exponent is yet another lexical item that must be memorized; the computational load is also increased, because the presence of the zero morpheme would have to be inferred by comparison with other entities and also because the zero morpheme might trigger operations such as readjustment rules. All in all, Siddiqi (2006) assumes that a grammar with fewer zero morphemes is more economical.

The notion of economy has been argued in different works to be a reason to make zero morphemes unavailable. Börjars and Donohue (2000) argue that economy, understood in a Minimalist sense (Radford, 1997, p. 505) as a constraint whereby linguistic representations should contain as few constituents as possible, makes zero morphemes at best severely restricted: under what conditions would a speaker posit that there is an additional constituent in the representation, without overt evidence for its presence? These authors point out that in other notions of economy, such as the LFG-based principle of Economy of Expression (Bresnan, 1995, p. 13), the plausibility that zero morphemes exist is even more reduced: LFG assumes that within a syntactic representation, only those nodes that are independently required are present. Many zero morphemes, as noted above, are proposed to produce regular and systematic segmentation within paradigms; according to Economy of Expression, the features that these morphemes could correspond to might not even be present in the structure.

Finally, some authors have proposed an acquisition constraint whereby linguistic properties can only be set through salient morphemes, where saliency among other things requires that the exponent is overt; this is a particular interpretation of Chomsky’s (2001, p. 2) Uniformity Principle—which in its usual interpretation does not ban phonologically null objects—whereby all variation should be acquired from easily detectable properties of utterances. While this type of claim is sometimes tacitly assumed, a few authors have made the point explicitly. Gervain and Mehler (2010, p. 194) present an acquisition model where the actual task that a human acquiring languages faces is to link abstract representations such as “noun” or “pronoun” to actual physical entities in the speech signal. Fasanella and Fortuny (2011) and in a more elaborate way Fasanella (2014, p. 24) state from this the Accessibility Condition in (18):

(18)

Leivada (2016, p. 116) reformulates this constraint as the PF-cues sensitivity principle: saliently accessible morphophonological cues are always required to set parameters during acquisition. If so, zero morphemes should either not exist or be used only for properties independently shown not to be subject to variation across languages, whatever those are.

Thus, beyond the methodological concerns, zero morphemes have been criticized because in principle they increase the memory load, go against the economy of representations, and make acquisition more difficult. Note, however, that the authors cited in this section do not claim that zero morphemes are completely impossible; for instance, the Accessibility Condition implies that no zero morpheme can be posited in order to account for a property that is subject to parametrization, but it does not exclude that they can be posited for properties that are assumed to be universal cross-linguistically and not subject to variation.

2.3 Theories That Allow Zero Morphemes, but Only Use Them Under Restricted Circumstances

The difficulty of banning zero morphemes completely has prompted other authors to allow them and use them, but with caveats directed to avoiding their proliferation.

Mel’čuk (2002) is perhaps the most influential author who has discussed overtly the principles that restrict the use of zeroes in morphology to avoid the methodological concerns that are standardly pointed out. He proposes four principles; in the specific case of zero morphology, E should be interpreted as a word form, and X as the alleged zero morpheme.

These rules severely restrict the use of zeroes. While the first principle would probably be accepted by anyone—do not posit a morpheme that is empty both in form and in meaning—the last principle makes zeroes unavailable when they are supposed to express properties not belonging to an obligatory morphological category in a paradigm. This excludes zeroes from derivational processes, for instance. The second principle makes zeroes unavailable in cases such as those that Nida (1948) criticized, namely when the word form undergoes any phonological process, including non-segmental ones, such as vowel changes, reduplications, etc. The third principle is meant to distinguish zero morphemes from cases where a word does not express a particular property: if the alleged zero does not contrast in the alleged position with a non-zero exponent, rather than claiming that there is a null morpheme, Mel’čuk proposes that the word is treated as invariable for that property. One relevant example is the English word in (19), where both singular and plural are unmarked.

(19)

Beyond the specific principles proposed, what is relevant for the purposes of this chapter is that even those who assume that zero morphemes are valid linguistic objects admit that methodologically they can become problematic, and therefore try to restrict their presence as last-resort analytical devices.

5.1 Positional Restrictions

One common property of theories that posit zero morphemes is the claim that a zero morpheme cannot be placed in any position within the word. Specifically, the restriction in (26) has been accepted by many as another device to restrict the distribution of zeroes in the analysis:

(26)

In some of the published literature, this is informally known as Myers’ Generalization (for instance, Bošković, 1997), after Scott Myers’s paper on zero derivation (1984). However, Myers didn’t state this as a principle, and in fact a particular interpretation of (26) is at odds with what Myers said. The generalization Myers identified is that once a form has changed category without an overt derivational affix, no further derivational affixes can be added to it. For instance, act is a verb that becomes a noun without an overt suffix. What Myers notes is that it is possible to overtly derive the verbal version (27a), but not the nominal version (27b).

(27)

However, this does not involve any claim about the position of zero morphemes, mainly because Myers interprets this restriction as meaning that no zero morpheme should be stipulated in the case of zero derivation. Instead, Myers’s claim is that the inflectional morphology is responsible for the category change in zero derivation: in (28) there is no zero nominalizer (as in 28a), and it is instead the inflectional number morphology that turns the structure into a noun (28b; cf. 1984, p. 55).

(28)

On the assumption that derivational morphology cannot follow inflectional morphology, his generalization is explained: to have zero derivation means that there is already an inflectional morpheme in the representation, and that structure cannot be further derived.

Nevertheless, the empirical side of the generalization has also been put into question. First of all, Bauer, Lieber, and Plag (2013, pp. 503–505) have shown that in many cases it is difficult to determine independently whether the base of a process is the noun or the verb. Second, Pesetsky (1995, p. 78) has noted that on the surface the generalization is violated at least by two English affixes: -er and -able.

(29)

Pesetsky (1995), who himself uses zero morphemes extensively in his work, concludes that there is no inherent positional restriction on such objects: any empirical generalization that restricts the combinations of zero morphemes and other derivational affixes must be studied through the individual lexical properties of the affixes involved.

5.2 Zero Morphemes and Inflectional Regularity

A second property that has been pointed out in the literature is more subtle, and in part theoretically conditioned by the assumption made in some theories that morphological irregularity is triggered by a surface morphophonological property. The idea, in very rough terms, is that whether a particular head is irregular is a property that has to be listed in the morphophonological representation of the head, not as a feature in its abstract morphosyntactic representation (see, for instance, Halle & Marantz, 1993; Don, 1993; Oltra-Massuet, 1999).

Assuming that property, if a zero morpheme is a morpheme that lacks morphophonological information, this means that a zero morpheme cannot be irregular, and whatever morphology combines with it and is selected by it will display the unmarked, elsewhere form.

This property can be illustrated through zero derivation in English. Don (2004) notes that most verbs ending in /ɪŋk/ or /ɪŋg/ in English are listed as irregular (30) in the past (2004, pp. 938–939).

(30)

The exceptions are those verbs that are zero-derived from nouns, such as ring in the sense of ‘to put a ring’ or ink, whose past tenses are ringed and inked, with the unmarked past tense morphemes. Don’s (2004) explanation is that being irregular as a verb—and thus combining with marked exponents for past tense—is a property that has to be listed in the verb itself. The case of the form ink has the structure of (31), and the verbal head is empty of morphophonological properties; therefore, it will not introduce an irregular exponent related to past tense, because that exponent for the category V does not carry with it the information to select the shape of other morphemes (32).

(31)

(32)

Remember, however, that the presence of zero morphemes is not disjoint with irregular morphology. Zero morphemes can trigger readjustment rules in the theories discussed in section 1. Their absence of morphophonological representation, then, prevents them from selecting marked forms in the outer layers, but not from triggering a morphophonological operation on the base.

5.3 Transparency for Morphophonological Selection

Another property of zero morphemes that has been proposed, and which also follows in principle from their lack of morphophonological information, is that zero morphemes are transparent for morphophonological selection. Embick (2010) has argued that in a situation like (33), where the sign ‘^’ stands for a linear adjacency relation, the morpheme A cannot directly select the shape of a morpheme C if the morpheme B is overt; however, if B is morphophonologically zero, A can condition the shape of C.

(33)

Let us illustrate this with the examples that Embick himself uses (2010, pp. 70–75), from a Latin conjugation. Note the subject agreement morphemes in (34a) as opposed to (34b).

(34)

The subject agreement forms in (34b) are the default morphophonological shapes, as they are found for instance in the imperfect indicative. Note that the perfect indicative subject agreement markers in (34a) are marked allomorphs in all forms except 3sg and 1pl.

(35)

Starting from these facts, Embick proposes that the Latin verb has the structure in (36), where perfect and imperfect are placed in Asp(ect) and the -ra- morpheme in the pluperfect is a past T(ense) form.

(36)

Starting from this analytical assumption, the argument is that the perfect -v(i)- exponent placed in Asp selects the set of special allomorphs for Agr in (34a), but it cannot do so when there is an overt morpheme intervening between Asp and Agr, as is the case with -ra- in T in the pluperfect. However, assuming that—irrespective of their surface manifestation—every verbal form in the paradigm contains at least these constituents, the perfect indicative must also contain in its subjacent structure a T head. However, the idea is that in that case this T head does not interrupt the selection between Asp and Agr because it is spelled out as a zero morpheme; specifically, Embick (2010, pp. 58–59) proposes that zero exponents are removed from the representation of any rule stating the linear adjacency relation between overt morphemes. The idea, again, is that the lack of morphophonological interpretation implies that the morpheme will not carry its own selection of marked allomorphs, or prevent other morphemes from introducing such allomorphs in the context.