1. Introduction
Null arguments are a common feature of written recipes, as long noted in the literature (e.g. Haegeman Reference Haegeman and Simon-Vandenbergen1987a, Reference Haegemanb, Massam & Roberge Reference Massam and Roberge1989, Massam Reference Massam1992, Cote Reference Cote1996, Culy Reference Culy1996, Bender Reference Bender1999, Ruppenhofer & Michaelis Reference Ruppenhofer and Michaelis2010, Ruda Reference Ruda2014, Weir Reference Weir2017). In (1) below, there is no overt agent for any of the verbs and the verbs cut and add are missing their patient argument.
In the context of recipes, the agent corresponds to the person following the recipe and the patient is what we will call the object of manipulation, which is the entity that is being acted upon and that undergoes changes throughout the creative process (Massam, Bamba & Murphy Reference Massam, Bamba and Murphy2017). Although the literature on recipes tends to refer to null subjects and null objects, we will adopt the terms ‘agent’ and ‘patient’ since thematic roles map differently to grammatical roles in different languages.
The literature on recipes has typically focused on the phenomenon of null definite patients, because these are generally ungrammatical in English. But as just noted, null agents are also found in recipe contexts. In this paper, we show that null agents and patients are a feature of recipes in a range of typologically and genetically diverse languages. Despite the fact that both arguments can be null, we argue that there is a fundamental asymmetry in the licensing of null agents and null patients. On the one hand, null agents in recipes are shown to be licensed via a range of different syntactic strategies across different languages, such as the imperative and the infinitive. On the other hand, we argue that that null patients in recipes are uniformly licensed when the null patient is bound by a null discourse topic merged in the left periphery. We will refer to this configuration as null topicalization.
As for the relation between syntax and register, our working hypothesis is that the recipe register does not specify particular syntactic structures (there is no universal recipe syntax) but has pragmatic desiderata.Footnote 2 In the case of recipes, there is a preference for null agents and null patients. Languages can satisfy these desiderata in different ways. Two questions arise: first, how do different languages meet these desiderata? second, how is the relation between register and syntax mediated? In this article, we focus on the first question, drawing on cross-linguistic data. As noted above, we argue for an asymmetry in how null arguments are licensed: null agents come about due to a variety of syntactic means, while null patients are the result of null topicalization. For the second question, we suggest that the recipe register makes salient the object of manipulation. It is this pragmatic salience that gives rise to the option of null topicalization, even in languages where null topicalization is highly restricted (e.g. English). Thus, although register and syntax remain independent, they do intersect. We discuss the interface between register and syntax in Section 5.
We note in passing that Weir (Reference Weir2017) and Haegeman (Reference Haegeman2017, Reference Haegeman2019) propose a unified analysis of null arguments found in recipes and in other instructional contexts, such as bottle labels (Sadock Reference Sadock, La Galy, Fox and Bruck1974), as well as those found in other reduced written registers such as diaries. We suspect, however, that the creative aspect of recipes, where the null patient corresponds to the object of manipulation that undergoes change leading to a specific output, makes them distinct from these other contexts. We therefore focus exclusively on null arguments in recipes.
This article is organized as follows. In Section 2, we examine null arguments in Malagasy recipes in detail. Section 3 then provides an overview of null agents and null patients in recipes from a range of languages, illustrating the variety of syntactic strategies deployed by different languages. We look more carefully at null patients in Section 4, where we present our null topicalization analysis. Section 5 considers the implications for the register–syntax interface and concludes.
2. Malagasy
Malagasy is an Austronesian language spoken in Madagascar (and in the diaspora) by over 25 million people. The basic word order is V(erb)O(bject)S(ubject). Important for this paper is what we will call the voice system: verbal morphology that serves to advance one argument to the clause-final position. This position has many different labels in the literature (e.g. subject, topic), but for the purposes of this paper, we follow Pearson (Reference Pearson2005) and call this position the trigger and assume that it appears in the specifier of a TopicP projection that dominates TP. VOS word order is obtained via predicate fronting, as proposed by Pearson (Reference Pearson2001, Reference Pearson2018), but the details of how word order is derived are not relevant for the purposes of this paper. In the examples below, the trigger is set in bold.Footnote 3
In (2a), the verb carries Actor TopicFootnote 4 morphology, and the agent (or highest argument) is the trigger (here i Bao). When the verb is marked for Theme Topic morphology, the trigger is a patient (ny akoho ‘the chicken’), as illustrated in (2b). Finally, there is Circumstantial Topic verb morphology, where some other element, such as an instrument or location, is the trigger. In (2c), the trigger is a benefactive (i Soa). When the agent is not the trigger, as in (2b) and (2c), it is realized adjacent to the verb, with what is called genitive case. We note that there are other voices, such as the a-passive (Keenan Reference Keenan and Li1976, Paul Reference Paul2000), as shown in (3) (so-called because it involves the prefix a-). The trigger of these clauses is the patient of some ditransitive verbs or the location of verbs like asiana ‘put’. In (3), for example, the trigger is ny latabatra fiaskako ‘my worktable’, the location where the flowers are placed.
As noted above, the agent of non-Actor Topic verbs can appear as a genitive phrase, right adjacent to the predicate, as in (2b,c), but it can also be omitted, as in (3). All of these aspects of the voice system will be important in what follows.
2.1 Malagasy recipes
The recipe data in this paper are mostly taken from a 1983 cookbook, Cuisine malgache, cuisine créole, by Pierre Boissard. Some additional examples were elicited from native speakers. Data source annotations set in brackets are detailed in the list of additional data sources, after References.
Like English, Malagasy has null agents and null patients in recipes. We see in (4a) below that neither verb has an overt agent and that the verb arotsaka ‘pour’ is missing its patient (for this verb, the patient is always the topic of the a-passive form). Similarly, in (4b), the agents and patients of ahandroina ‘cook’ and esorina ‘remove’ are null.Footnote 5
We note here that Malagasy recipes do not use the imperative (unlike English). Malagasy has overt imperative morphology: the imperative forms for the verbs in (4a) would be sasao ‘wash!’ and arotsahy ‘pour!’, respectively. Moreover, Malagasy lacks a dedicated infinitive form, so we assume these verbs are not infinitives. Instead, what is striking about the verbal morphology in recipes is that it is typically non-Actor Topic, whether Theme Topic, Circumstantial Topic, or a-passive (see Keenan & Manorohanta Reference Keenan and Manorohanta2001 for a discussion of the prevalence of non-Actor Topic forms in Malagasy texts). The question that now arises is how null agents and null patients are licensed in recipe contexts in Malagasy.
2.2 Null agents
As noted above (see example (3)), null agents are always possible with non-Actor Topic verbs (much like agents in English passive). We can see a null agent in the example below, where the verb hosorana ‘smear’ is Theme Topic and ny volo ‘the hair’ is the trigger.
Much like null agents in English passives, the null agent here is interpreted as an indefinite or generic (‘someone smeared pomade on the hair’). Given that most verbs in Malagasy recipes are in the non-Actor Topic form, null agents will always be possible.Footnote 6 No special licensing conditions are required.
2.3 Null patients
As just noted, most verbs in recipes are in non-Actor Topic forms, and mainly Theme Topic. Recall that when the verb is Theme Topic, the patient is in the trigger position. We illustrate with the example in (4), repeated here as (6). In (6a), the verb sasana ‘wash’ is Theme Topic and the patient, ny vary ‘the rice’ is the trigger. In the subsequent clause (6b), the verb arotsaka ‘pour’ is in the a-passive form, so the trigger is the patient (the rice), which is null.
In what follows, we argue that null patients arise due to an independently available phenomenon in the language, referred to as trigger-drop (or topic-drop) in the literature (Keenan Reference Keenan and Li1976, Randriamasimanana Reference Randriamasimanana1986, Potsdam & Polinsky Reference Potsdam and Polinsky2007). We provide examples in (7).
Potsdam & Polinsky (Reference Potsdam and Polinsky2007) argue that the null arguments in (7) are pro rather than PRO. They propose that pro is licensed in Spec, TopP by Top0 (recall that this is the position that triggers occupy in Malagasy) and that pro is identified via coindexation with the current discourse topic. Trigger-drop is possible in both root and embedded clauses, as illustrated in (7). Departing slightly from Potsdam & Polinsky (Reference Potsdam and Polinsky2007), we suggest that the null trigger is bound by a null discourse topic that is merged in the left periphery (similar to Huang’s Reference Huang1984, Reference Huang and Freidin1991 proposals for null objects in Mandarin Chinese – see Section 4 for more discussion).Footnote 7 If an overt trigger is present, such as Rabe in (7a), the trigger will preferentially be understood as being the same as the discourse topic (see Potsdam & Polinsky Reference Potsdam and Polinsky2007: 292–296 for some discussion). Thus, while it may appear that the null trigger is bound by the overt trigger, we claim that the null discourse topic is the binder.
Instances of trigger-drop can easily be found in written texts, such as folk tales. In (8a), for example, the first clause contains the trigger, ireo lefona ireo ‘those spears’. We suggest that a null discourse topic, coreferent with the preceding trigger, is merged in the left periphery of the second conjunct and binds the null trigger. Note here that the predicate natorana ‘throw’ in the second conjunct is in the a-passive form and therefore the null trigger is the missing patient of this verb (the spears in this instance). In (8b), on the other hand, the null trigger of the adjunct clause is bound by a discourse topic that does not correspond to the trigger of the main clause (ianao 2sg). Instead, the discourse topic is understood to be peratra ity ‘this ring’.
In other examples, there is no overt DP that corresponds to the discourse topic, much like the root clause in (7b). For example, in (9a) the missing trigger of nodinihiny ‘examine’ is a carpet, and in (9b), the missing trigger of vonoy ‘kill’ is a man, both previously mentioned in the story.Footnote 8
Similar facts hold in recipes: the antecedent of the null trigger is always the current discourse topic. Unlike in other texts, however, in the context of recipes the discourse topic is fixed: it is always the object of manipulation. Like the folk tale example in (8a), the discourse topic in a recipe can be overtly expressed via a trigger. In (10a), for example, the object of manipulation is the meat (which corresponds to the trigger of the first clause), and it is this discourse topic that binds the null trigger in the second clause. Much like we saw in (9), there is not always an overt DP that corresponds to the discourse topic. In (10b), there is a null trigger in the first clause that is understood to be the location of the action of putting, as signalled by the a-passive morphology. The antecedent is the object of manipulation; in this case, the soup that the salt is being added to.
While the discourse topic can correspond to an overtly realized trigger (e.g. ny hena ‘the meat’ in (10a)), it can’t correspond to most other syntactic positions. In (11), for example, the trigger of the first conjunct is ny tahon’anana ‘the vegetable stems’ but in the context of the second conjunct, this leads to a pragmatically dispreferred interpretation where the stems are put in the pot (rather than the leaves). In other words, the discourse topic cannot be understood as the possessor anana ‘vegetables’.
Note that the English translation of (11) has a similar interpretation. We suggest that anana ‘vegetable’ has been backgrounded (as a possessor in Malagasy or in a compound in English) and therefore cannot correspond to the discourse topic.
Finally, we note that the discourse topic does not have to correspond to the overt trigger, as in (12) below (see also (9b) above). Here the trigger in the main clause is ny herin’ny afo ‘the intensity of the flame’ but the discourse topic is the object of manipulation, the soup that the salt is being added to.
In other words, although there is an overt trigger in (12) (ny herin’ny afo ‘the intensity of the flame’), the null trigger in the second clause is bound by the discourse topic (the object of manipulation, the soup). Thus, the null patient can correspond to a previous trigger, as in (8a) and (10a), or to a discourse topic, as in (9), (10b) and (12).
Summing up, Malagasy recipes have null agents and null patients. Null agents are possible due to non-Actor Topic voice morphology, which independently licenses null agents. Null patients arise due to trigger-drop, a widespread phenomenon in the language. We note in passing that both properties rely on non-Actor Topic voice, predicting that null arguments should not be possible with Actor Topic verbs. In this context, the few instances of Actor Topic in the recipe book were revealing. The verb is always mangotraka ‘boil’, which is an unaccusative verb, lacking an agent. Given that an overt agent is not possible, the trigger is therefore the highest argument (here the patient), and it can undergo trigger-drop. In other words, because mangotraka ‘boil’ is unaccusative, it patterns with non-Actor Topic verbs in allowing trigger-drop of the patient.
Complicating the picture is the fact that manongotra ‘boil’ is embedded under the matrix predicate avela ‘let’. A more careful study of other voice forms in recipes is left to future research.
What Malagasy recipes have shown is that null agents and null patients are attested in recipes contexts outside of English, but the question arises as to whether the licensing conditions for these null arguments differ across languages. To better understand the cross-linguistic variation, we now turn to null arguments in recipe contexts in a range of languages.
3. Null agents and patients in other languages
We have just seen how Malagasy licenses null agents and null patients in recipes, via non-Actor Topic morphology and trigger-drop, respectively. This section investigates other languages to show that a range of syntactic strategies are used cross-linguistically. We note here that our language sample is based on convenience but does include languages from different language families and with different typological properties. Moreover, our discussion of recipes in these languages should not be taken as exhaustive, but merely illustrative. For example, French recipes may appear in either the infinitive or the imperative.
3.1 Null agents in other languages
As we saw at the start of the paper, English recipes use the imperative mood, where agents (subjects) are typically omitted.Footnote 9 In fact, Cotter (Reference Cotter and Bower1997) considers the imperative to be the recipe’s most distinguishing feature (see Fischer Reference Fischer2013, also Fisher Reference Fisher1983). Moreover, imperatives have been used in recipes since at least Middle English (Arendholz et al. Reference Arendholz, Bublitz, Kirner-Ludwig and Zimmerman2013). An illustrative example is given in (14), where the agent of sift is null.Footnote 10
Since the syntax of these null agents is presumed to be identical to the syntax of imperatives, null agents in recipes have not received much attention in the literature.Footnote 11
Looking at other languages, both Niuean and Tagalog also use imperatives in recipes, as shown in (15)–(17). As seen for Niuean in (15), there is no overt agent for either verb (helehele ‘slice’ and kai ‘eat’).
While imperatives are not morphologically marked in Niuean, there is a special form of negation (ua) that is only used for imperatives (Seiter Reference Seiter1980), and it also occurs in recipes (16a) below. This negation is distinct from the sentential negation nākai, seen in (16b). We take this distribution to show that recipes indeed use the imperative in Niuean.
A similar situation obtains in Tagalog, as in (17).Footnote 12
While the imperative is not overtly marked (imperatives are aspectless), there is a special form of the negation (huwag) that only occurs with imperatives and is also used in recipes (18a). The example in (18b) illustrates sentential negation (hindi) (data and glossing from Henrison Hsieh p.c.).
Not all languages use the imperative in recipes, however. We have seen that Malagasy does not, and French and German recipes can appear in the infinitive, as illustrated in (19) and (20), respectively.
Notably, the infinitive is another syntactic structure that typically lacks an agent (subject).
Null agents can also arise due to pro-drop, as in Japanese, where recipes do not use the imperative; instead, the verb in marked with the conclusive form (a finite form that concludes a sentence and is not otherwise necessarily associated with a null agent) (Shimojo Reference Shimojo2019).
Hinds (Reference Hinds1976) notes that agents are always null in Japanese recipes. Since null agents are licensed in general in Japanese, via radical, or discourse, pro-drop, we can posit that this mechanism is also available in recipes.
Finally, Bulgarian recipes use middles for recipes (among other strategies; Vesela Simeonova, p.c.), as seen in (22).
In middle constructions, like imperatives and infinitives, agents are normally excluded.
Summing up, we claim that the recipe register dictates that the agent is the person following the recipe, and that due to its pragmatically given identity, the agent is preferably null. Syntax operates on this directive via different means (imperative, infinitive, pro-drop, etc.), depending on the language.
3.2 Null patients in other languages
Just as we saw for agents, null patients are allowed in recipe contexts in all the languages we looked at. The possibility of null definite patients has been a puzzle for English, where such null arguments are typically not possible. Consider the contrast in (23), where (23a) is well-formed in the context of a recipe, but (23b) is not a recipe and the null argument of boil leads to ungrammaticality.Footnote 13
Many authors have addressed this issue (e.g. Haegeman Reference Haegeman and Simon-Vandenbergen1987a,Reference Haegemanb, Massam & Roberge Reference Massam and Roberge1989, Massam Reference Massam1992, Ruda Reference Ruda2014, Massam et al. Reference Massam, Bamba and Murphy2017, Weir Reference Weir2017). The details of their analyses differ, but for the most part, the differences between them relate to the nature of the null object (whether it be a trace, a DP, D, or nP, for example). With respect to the recoverability of the null object, however, all these authors consider that the null element must be bound by a null discourse-determined antecedent (referred to as a ‘running topic’ by Massam et al. Reference Massam, Bamba and Murphy2017).Footnote 14 We assume that the same analysis applies to French. As seen in (24), repeated from (19), the patients of couvrir ‘cover’ and cuire ‘cook’ are null.
The question then arises as to why this null patient is possible in English (and French) in recipe contexts like (23a), but not in (23b). We provide a more detailed discussion of null patients in English and more generally in the next section. The remainder of this section is devoted to null patients in recipes in the languages discussed in the preceding section.
In some languages, as we saw for Malagasy, the null patient arises due to a process of trigger-drop, which is found in other sentence types as well, whereby a null argument is bound by a null topic in the left periphery. A similar phenomenon occurs in Tagalog recipes. In (25), the verbs are in the Goal Topic voice, so the missing patients are the trigger of their proposition.
In some languages, pro-drop allows for the omission of the patient. For example, Bulgarian is a pro-drop language and the patient can be dropped in (26b). Recall that Bulgarian uses middles in recipes so the omitted patient (‘the onion’) is the subject of the verb zadushava ‘sauté’.
On the other hand, Japanese is a radical (or discourse-licensed) pro-drop language, so we can assume that the null patient (‘the chicken’) in (27b) arises via pro-drop.
Finally, Niuean is also a radical pro-drop language, so we consider it to be like Japanese. We note in passing that there is no overt form for third person inanimate pronouns (and most, if not all, objects of manipulation are inanimate). Such pronouns are therefore obligatorily null. We can see this in the examples below. The example in (28a) illustrates that there is no overt correlate to the English pronoun it. The pronoun is syntactically present, we claim, as there is ergative case marking – a clear signal of transitivity. The recipe example in (28b) (repeated from (15)) could simply have the same null inanimate pronoun (Massam et al. Reference Massam, Bamba and Murphy2017).Footnote 15
While the data are open to both interpretations, we assume in what follows that Niuean recipe null objects are licensed by the mechanism of optional pro-drop found elsewhere in this language.
Summing up, we have claimed that the recipe register dictates that the patient is the object of manipulation and that the patient is preferably null. It seems that with null patients, just as with agents, the syntax of individual languages operates on this directive in different ways, such as by trigger-drop, pro-drop, or via a running topic. Combined with the observations about null agents in the previous section, the picture in Table 1 emerges.Footnote 16
Syntactic strategies for null agents and null patients – to be revised.
As can be seen in Table 1, different languages have different syntactic strategies, which fits with our working hypothesis that there is no ‘recipe syntax’, per se. In the next section, however, we look more closely at null patients and consider some cross-linguistic similarities, which will lead us to a different conclusion: while null agents are licensed in a range of ways, null patients are uniformly licensed by null topicalization. The implications for the relation between syntax and register are discussed in Section 5.
4. More on null patients
As we saw in the previous section, it is clear that different languages use different syntactic resources to license null agents (imperative, infinitive, voice, etc.). At first glance, this also appears to be true for null patients. However, null topicalization, in which both the nominal in the argument position, and the topic in the left periphery are null, turns up in English (and French) and Malagasy (and Tagalog). We therefore now ask whether null topicalization could also account for null patients in recipes in radical pro-drop languages, such as Japanese and Niuean.
4.1 Null patients and null topicalization
As already noted, topicalization typically involves two entities, one in the argument position, and another in the left periphery. In more familiar cases, the former is null and the latter is overt, as in (29).
It has been proposed, however, that in some cases, the left peripheral topic can also be null, as we saw in Malagasy, a phenomenon we refer to as null topicalization. The connection between the possibility for null patients in a language and the option for null topicalization goes back at least to Huang (Reference Huang1984, Reference Huang and Freidin1991). Huang argues that null objects in Mandarin Chinese, a radical pro-drop language, are bound by a null topic in the left periphery. Thus, the null object of renshi ‘know’ in (30) is bound by TOP. He notes that what is special in this Mandarin sentence, in contrast to English, for example, is not that the object is null, since English topic-bound objects are also null, but that the topic is null, which is allowed in Mandarin, but not in English, due to the discourse-oriented nature of the language.Footnote 17
Null topics have also been argued to exist in several other languages, such as Malagasy, discussed above, and also European Portuguese, German, Russian, and Hebrew (e.g. Raposo Reference Raposo, Jaeggli and Silva-Corvalán1986, Erteschik-Shir, Ibnbari & Taube Reference Erteschik-Shir, Ibnbari and Taube2013, Trutkowski Reference Trutkowski2016). These authors propose topic-drop to account for the existence of null objects in these languages, and this analysis is supported by the fact that such null objects are always topical.Footnote 18 In German, for example, for an object to be null, its reference must be given by the discourse context and it must be salient, that is, more prominent at a point of time than other units of information (Trutkowski Reference Trutkowski2016). Adapting this proposal to recipe contexts is initially appealing as recipes by their nature always have a clear discourse topic – the object of manipulation. It is possible, therefore, to posit that across languages, null objects in recipes are bound by null topics. The proposal marries radical pro-drop (Barbosa Reference Barbosa2011) and recipe null objects, as the latter now effectively exhibit register-determined radical pro-drop, to the extent that radical pro-drop involves null topicalization (Massam et al. Reference Massam, Bamba and Murphy2017).Footnote 19
In the literature on null topicalization, it has been observed that languages can vary in terms of what conditions license it, with Mandarin and Japanese allowing a wider range of options than German and Russian (Saito Reference Saito2007, Erteschik-Shir et al. Reference Erteschik-Shir, Ibnbari and Taube2013).Footnote 20 Looking at English, for example, null topics are disallowed in contexts where they are permitted in German. Thus, in answer to the question in (31a), (31b) is ungrammatical, in contrast to the well-formed German example in (32b).
However, it is not the case that English altogether disallows null objects that are contextually given and highly salient. Rather, the licensing conditions for null topics are tighter than in some other languages, ruling out linguistic discourse topics (e.g. (31b)) and allowing only null topics that are directly discernable from deixis or extra-linguistic context (Noailly Reference Noailly1997; Cummins & Roberge Reference Cummins, Roberge, Julie Auger and Vance2004, Reference Cummins and Roberge2005; Perez-Leroux et al. Reference Perez-Leroux, Pirvulescu and Roberge2017). The latter option is also noted for Russian and Hebrew by Erteschik-Shir et al. (Reference Erteschik-Shir, Ibnbari and Taube2013). The examples in (33) illustrate instances where the null object’s reference is immediately salient in the extra-linguistic context (see Cummins & Roberge Reference Cummins and Roberge2005 for similar examples).Footnote 21
In both examples in (33), it is clear that there is a null definite object. In particular, in (33b), the parent is telling the child to eat the broccoli, not just to eat something. Assuming that such objects are contextually topical, we can conclude that languages differ in the extent to which they allow null topicalization, with licensing conditions varying across pragmatic and discourse lines. This means that we need a more finely tuned approach to cross-linguistic options for null topicalization (Erteschik-Shir et al. Reference Erteschik-Shir, Ibnbari and Taube2013). We propose that while languages can have either more or less expansive conditions for null topicalization, recipes are always within the boundaries of admissibility, across (all) languages.Footnote 22 Therefore, a language like English, with very particular tight constraints on null topicalization, nonetheless allows it in the recipe context.
There is an interesting correlation between the conditions on null topicalization in non-recipe contexts and in recipe contexts. Schulz (Reference Schulz, Ikeda, Robideau, Ballantyne, Garneau, Hall and Lanz2003) notes that null topicalization is not available for all types of topics, for example shifted topics, but is used only for continued topics.Footnote 23 This difference has also been observed in Japanese recipes by Shimojo (Reference Shimojo2019). Japanese recipes can use overt topics: they are used to introduce new ingredients (e.g. toriniku-wa ‘chicken’ in (34a)). Recipes can also deploy null topics (e.g. ‘the chicken’ in (34b)) to bind null patients, but in contrast to the overt topics, the null topics are the continued ingredients and are used in what Shimojo calls series cohesion (similar to null anaphora in texts), where the same entity (the continued topic) is manipulated across an ongoing set of instructions.Footnote 24
Thus, overt and null topics play different roles in the discourse, both in regular null topic contexts and in recipe null topic constructions, with null topicalization being employed in cases of continued or series cohesion topics.
As we noted above, Niuean is another language that allows null objects more broadly than English, but Massam (Reference Massam2020) argues that null objects in Niuean are in fact found in very specific contexts. Once such context is where the object is highly topical, with a close linguistic antecedent in an immediately local, but syntactically separate, sentence, as in the two examples below.Footnote 25
If recipes constitute a cross-linguistically discourse-licensed null topic context, then Niuean recipe null objects such as in (15) are the same as those in non-recipe contexts in the language such as (35) and (36), which are bound by null topics.
We thus see that across languages we find null patients that have been argued to be bound by null topics in the left periphery. These null left-peripheral topics might be licensed (i.e. achieve referential identity) very restrictively, by means of the extra-linguistic context (as in English), or they might be licensed by discourse topics from the preceding discourse (as in Mandarin Chinese, German, among others), with varying conditions on locality, degree of topicality, and so on. Based on this previously discussed phenomenon, our claim here is that recipe context sentences are provided with topics by virtue of their register, and that these topics are permitted to be null across (all) languages because of their tightly defined pragmatic and discourse topicality. We return to a discussion of the interface between register and syntax in Section 5, but we now look more closely at the syntax of null topicalization.
4.2 How null topicalization works
In this subsection, we define the notion of topic that we claim is relevant for the proposed null topicalization and we illustrate how this phenomenon applies in different languages.
We adopt the core definitions of the different kinds of topic found in Frascarelli & Hinterhölzl (Reference Frascarelli, Hinterhölzl, Schwade and Winkler2007). Crucial for recipes are familiar topics, which are given, are typically destressed and pronominal, and are involved in topic continuity. We claim that in recipes, there is a null familiar topic in the left periphery of the clause that binds a null argument. Note that while there is some debate in the literature on the availability of topics in imperatives (Zhang Reference Zhang1990, Barbiers Reference Barbiers2007, Koopman Reference Koopman2007, among others), Frascarelli & Jiménez-Fernández (published online 21 March Reference Frascarelli and Jiménez-Fernández2021) show that some types of topics (given and contrastive) are permitted in imperatives, with some interlinguistic variation (see also Haegeman Reference Haegeman2012). As noted in Section 4.1, the constraints on this null topic binding vary among languages, but the recipe context always licenses this null topic. Moreover, the null topic always corresponds to the object of manipulation. The exact syntactic position of the argument is not fixed, rather it is the thematic role of patient and the syntax of recipes in a given language will determine the syntactic position of the patient. We illustrate below.
In English, recipes are in the imperative and as a result the patient is always mapped to the direct object, as in (37) (repeated with minor modifications from (1) above).
The proposed structure of the second sentence is represented as in (38), where TOP corresponds to the two carrots introduced in the previous sentence:Footnote 26
The same representation can be extended to French recipes, as well as recipes in radical pro-drop languages like Japanese and Niuean.
In Malagasy and Tagalog, however, the patient is not in the direct object position, but in the trigger position. We nevertheless posit the same analysis: a null topic in the left periphery of the second clause of (39) binds the null patient of the verb sasana ‘wash’. This verb is in the Theme Topic voice and therefore the patient is the trigger.
Again, we assume that the null familiar topic is not a root phenomenon and can appear in the left periphery of embedded and adjunct clauses. Familiar topics have independently been argued to not be a root phenomenon (Bianchi & Frascarelli Reference Bianchi and Frascarelli2010, Jiménez-Fernández & Miyagawa Reference Jiménez-Fernández and Miyagawa2014, Jiménez-Fernández Reference Jiménez-Fernández2020, inter alia).
For languages like Bulgarian, the precise details depend on the analysis of middles. In other words, the null topic could bind a null patient in the subject position, as in (40a), much like we saw for Malagasy. On the other hand, the topic could directly bind the null patient in its merge position (as object), as in (40b), along the lines of the English example in (38).
What unites these examples is that there is binding of a null patient by a null topic in the left periphery. The null patient can be in different syntactic positions, depending on the language. The register therefore plays an important role here in constraining what the null topic can bind. Outside of recipes, familiar topics are not normally constrained to a particular thematic role. But in recipes, the familiar topic is always the object of manipulation and therefore the patient.
4.3 Summary
We have suggested here that patient drop arises due to null topicalization, following the work of others cited above, such as Huang (Reference Huang and Freidin1991) and Erteschik-Shir et al. (Reference Erteschik-Shir, Ibnbari and Taube2013). We have also seen that languages differ in terms of what constraints exist regarding the licensing of null topics. Some languages allow null topicalization more freely than others. However, we have claimed that the recipe context always licenses null patients via null topicalization cross-linguistically. In other words, recipe context null patients are always licensed by null topicalization.
We note in passing that we do not address the categorial nature of the null patient in a recipe. Both Ruda (Reference Ruda2014) and Weir (Reference Weir2017) propose that it is a D-less nominal, on analogy with East Asian languages that exhibit discourse or radical pro-drop, and lack determiners (Tomioka Reference Tomioka, Schwabe and Winkler2003).Footnote 27 However, as argued by Massam et al. (Reference Massam, Bamba and Murphy2017), Niuean is also a radical pro-drop language, but it arguably does not lack determiners. We tacitly adopt Massam et al.’s view, where the null object is a D, that is, a simple deictic pointer, definite but without identifying content, which it receives from its antecedent (Roberts & Holmberg Reference Roberts, Holmberg, Biberauer, Holmberg, Roberts and Sheehan2010, Baltin Reference Baltin2012). We leave this issue aside here, as our main point involves the licensing of the null topic, rather than the categorial nature of the null patient itself.Footnote 28
5. Conclusion
Most of the research on recipe contexts tends to focus on one language and one issue, for example on how to explain null definite objects in English. By taking a cross-linguistic perspective, we see that instances of recipes in a range of languages share two key properties: null agents and null patients. We can understand this nullness functionally. The null agent corresponds to the reader, that is, the person following the recipe. There is no need to make this argument overt. The null patient is the object of manipulation and as a result, it is highly salient and can be null. But the functional account doesn’t tell us how any given language will make these null arguments possible in the syntax.
The languages explored in this paper show that different strategies are used by different languages. As can be seen in the table below (revised from Table 1), null agents arise due to a variety of syntactic means that exist independently in the language. In the case of null patients, however, we have claimed that they are always the result of null topicalization. Table 2 improves on Table 1 by capturing the generalization that the apparently different processes that lead to null patients (e.g. trigger-drop, pro-drop) all involve the same syntactic structure: the null argument is bound by a null topic in the left periphery.
Syntactic strategies for null agents and null patients – revised.
The emerging picture for agents conforms straightforwardly to our initial hypothesis that register does not dictate specific syntactic structures. Instead, the syntax of each language plays a role in realizing the pragmatic desiderata of the register. For null patients, however, several questions arise regarding the precise relation between the register and syntax. Does the recipe register provide the null topic that licenses the null patient? If yes, then we are committed to the view that a register can directly license a particular syntactic configuration (see Bender Reference Bender1999). Alternatively, it is possible that the salience of the topic in recipe contexts is so strong that it fits into every language’s allowable space for the licensing of null topicalization. We have seen that in all languages considered, even English, null topicalization is an option, given the right discourse or pragmatic context. Crucially, the licensing of null patients involves an independently available syntactic construction, not a mechanism only found in recipes. If this is the case, then we do not need a direct link between the recipe register and syntax. While we contend that the second approach is preferable, we leave it to future research to determine which approach to register and syntax is ultimately correct.
