The French verb laisser, much like its English counterpart let, can realise syntactic causative constructions, i.e. constructions in which both the causing event (e₁) and the caused event (e₂) are projected syntactically. In Romance syntactic causatives, e₁ is canonically expressed through a light verb, while e₂ is encoded by a non-finite clause embedded under said light verb. Those constructions have received a lot of attention in the literature for their idiosyncratic properties; however, much of that work focuses on the light verb faire and its Italian cognate fare, both ‘to make’ (Kayne 1975; Guasti 1996; Folli & Harley 2007; Ciutescu 2019). In Faire-Infinitive constructions (FIs), the light verb combines with the non-finite verb, hence creating a complex predicate. Complex predication occurs when two simplex predicates — one of them a light verb — combine in order to form a single complex one with a single argument structure (Butt 1995; 2010).1 The complex predicate faire chanter has two arguments: a Causer (the subject of the complex predicate) and a Causee (its object). This differs from their English counterpart, in which the Causee is the subject of the embedded verb.

Surprisingly, let-verbs across Romance have received less attention for their syntactic properties, despite realising (at least) two distinct syntactic causative constructions: one of them of type FI, as in (3a), the other realising an Exceptional Case Marking construction (ECM hereafter).

The existence of these two constructions led authors such as Borel (1972); Kayne (1975); Silva (2012); Enghels & Roegiest (2012; 2014) to assume that each construction possesses distinct semantic properties, potentially related to their (also distinct) morphosyntactic properties. Notably, Kayne (1975: 222) uses the contrast between the pair of sentences below in (4) to propose that the FI construction is unmarked, while the ECM construction would carry an intentional flavour (hence a situation in which the guard acted with ‘complicity, or deliberate neglect’).

On the other hand, Silva (2012), and Enghels & Roegiest (2012; 2014) propose a more fine-grained analysis, in which French laisser, Spanish dejar, and Portuguese deixar can accommodate three main readings: (i) permitting, (ii) not opposing, and (iii) releasing.2 As noted by Raffy & Donazzan & von Heusinger (2024), the release reading is not relevant for French laisser, as French possesses another verb, lâcher, with which laisser shares the Latin root verb laxare. Therefore, I will be using the terminology of Raffy (2021): (i) authorise, and (ii) not-intervene, which I discuss in section 2.2.2. These readings need not be tied to a specific structure, but Enghels & Roegiest (2012) do observe that the FI construction is more common than the ECM in the French corpus FRANTEXT. Yet, it appears that the FI is dispreferred with animate Causees. Consequently, if we assume that authorising requires two volitional (hence animate) entities in order to be felicitous, then the authorise reading is likely to be encoded by an ECM construction rather than an FI. While this does not seem to suggest as strong a mapping as that proposed by Borel (1972) or Kayne (1975), it still seems to hint at some degree of preference for one construction over the other for each given reading. I propose that these two readings are both non-interference readings, only with distinct aspectual properties which are due to the properties ascribed to the Causer. These differences are developed in the next section.

While counterfactual approaches to causation view it as a necessary relation between events, force-theoretical approaches to causation view it as a relation between (two or more) entities having intrinsic tendencies; how these entities interact with one another is what defines the causal relation (Talmy 1988). In force theories, letting (as well as related notions such as helping and enabling) is generally defined by opposition with causing. Causing refers to situations in which there is (a) opposition between Causer (Affector) and Causee (Patient) as (b) the Causer has a tendency for p while the Causee has a tendency either for ¬p or for neither p nor ¬p. That tendency of the Causee is then overcome by the Causer’s, hence leading to (c) the occurrence of a result. Letting, on the other hand, is analysed by Talmy (1988) as encoding reverse force interactions in which the Causee has a tendency for p, which is blocked by the Causer’s, who has an tendency towards ¬p. The opposition exerted by the Causer is then removed, hence allowing the result to occur. In a later approach, Wolff & Song (2003); Wolff & Thorstad (2017) propose that the notion of opposition is actually not necessary to the understanding of the notion of enable (under which Talmy’s letting is expected to fall). The characteristics for each causal notions as described in Wolff & Song (2003) are replicated in Table 1 below:

The English verb let, when it enters syntactic causative constructions, is generally understood as encoding relations belonging to the causal notion of enable. For this reason, it is often interchangeable with other enable-verbs like allow.7

At first glance, the criteria appear to align well with let: it is reasonable to infer that the art teacher allowed the action because (i) Peter had the intention to paint his own portrait, and (ii) the teacher had no objection to it. Given that both the Causer and the Causee displayed tendencies favourable to the result, the occurrence of the result is a predictable. If laisser-causatives were to be interpreted as enable situations, the three inferences in (17) would be true if (16) were true.

But, in fact, this is not quite the picture that we get from the interpretation of laisser-causatives. Instead, a sentence like (16) yields two possible sets of inferences: if it is read as an authorise scenario, then it triggers the inferences in (18), and if it is interpreted as a not-intervene situation, we get the inferences in (19).

When comparing (17) with (18) and (19), we can see that the first inference remains the same, therefore enable and let share one core principle. However, the second one already differs: while (17b) indicates that Mar intends for David to sing, (18b) and (19b) make no such claims. Instead, Mar is indifferent to David’s singing; in other words, she does not care whether he sings or not. The non-opposition component remains, but Mar’s influence over David’s singing is to be understood as neither stimulatory nor inhibitory. This means she does not push him (force or encourage) him to sing, and neither does she prevent him from singing. The last point ought to be discussed, as this is where authorise and not-intervene diverge; while (18c) highlights that the result does not necessarily occur in authorise-relations, (19c) shows that the so-called result is already ongoing in not-intervene relations (which seems like it goes without saying, but must be pointed out nonetheless). Thus, the way the would-be prevention is exerted slightly differs. In authorise-contexts, David’s choice to sing or not remains open, because his singing is not influenced by any external force (e.g Mar’s intention). He can either choose to sing or not to sing, depending on his intention at t. On the other hand, the would-be prevention is more subtle in not-intervene relations, as the singing is already occurring. Nevertheless, the fact that the situation is expressed through a laisser-causative in which Mar receives the role of Causer is taken to mean that Mar is perceived as being able to prevent the singing if she were to choose to. Canonical force-theoretical theories such as that of Wolff & Song (2003) do not seem to provide us with an guide for understanding the negation of a concept; and yet, this is what we are interested in here – what does it mean to not prevent? Is it the same as authorising? Apparently not, at least not when lexicalised through a laisser-causative. And since those theories fail to account for situations in which the Causer’s tendency cannot be defined in terms of being for or against the result, then they are too restrictive to include concepts such as let.

The claim that laisser does not encode enable-relations is further supported by the fact that it does not pattern with permettre ‘to allow’ the way their English counterparts do. The contrast is particularly salient when the enabling relation is made explicit, as in (20a):

This inability to encode enable-relations indicates that laisser must have a semantics distinct from that of let. It thus remains challenging to determine precisely what laisser contributes to our ontology of causation. The next section examines the limited involvement of the Causer in laisser-causatives and emphasises the importance of viewing causative constructions as linguistic representations of perceived causal relations, rather than as direct reflections of objective or ‘true’ causation.

Although it has now been established that laisser is not quite an enable verb, the question of its definition remains. The previous section introduces the idea that the Causer’s involvement is too minimal to qualify for enable; the logical next step is to define the Causer’s participation to the causal relation. In normality approaches to causation, participants in a causal chain are classified as either causes or enabling conditions, depending on whether they meet criteria of necessity or sufficiency (Mackie 1965; 1974; Hart & Honoré 1985; Cheng & Novick 1991; 1992; Wolff & Song & Driscoll 2002; Nadathur & Lauer 2020). As defined by Verschueren et al. (2004), a necessary condition is one without which the outcome cannot occur, while a sufficient condition guarantees the outcome. Let us consider our example with Mar and David above in (16), repeated below in (21).

If laisser-causatives stood for enable-relations, they would be able to select enabling conditions as Causers. Enabling conditions are causal factors that are insufficient but necessary for the result, they are also considered to be both normal and stable (Cheng & Novick 1991). Is Mar’s letting necessary for David’s singing? When no letting is performed, the result may still occur, as is exemplified by (22), which ought to be interpreted as meaning ‘even though Mar did not explicitly state her agreement for David’s singing, David still sung’.

Additionally, the discussion on causation by omission in section 2.2 highlights that the Causer, being a would-be preventer, fails to act normally. However, Mar’s letting is not sufficient either for the occurrence of the result; in authorise-relations, the Causee is not bound to sing by the letting and she can still choose not to sing if it agrees with her tendency (see discussion about (10) in section 2.2.2). For this reason, let-relations allow for a non-culminating reading of the caused event, as in (23):

In this light, the would-be preventer in cases of omission, such as Mar in the let readings, is minimally involved and appears to satisfy neither necessity nor sufficiency, further complicating its classification within standard causal frameworks. If the Causer of laisser-causatives is both insufficient and unnecessary for the result, how come they can enter causal relations at all? If they had no causal relevance, one could expect (24a) and (24b) to have the same meaning:

The reason why they differ is because the speaker attributes the role of would-be preventer to Mar, because she believes Mar could or should have prevented David from singing, but simply does not because she does not care for the outcome. This raises an issue that cannot be made visible by approaches that do not treat causation by omission: the speaker’s knowledge or evaluation of the properties (or assumed properties) of the Causer and the Causee is relevant in understanding the expression of causal relations in natural languages. In other words, linguistic expressions of causal relations are often more of a reflection of the speaker’s knowledge and beliefs rather than a true representation of actual causes and effects following the laws of physics. One famous example of the way our belief system and moral compass influences our judgment of causal relations is the Knobe effect, or side-effect effect (Knobe 2003; 2004). The Knobe effect shows that the bringing about of side effects of actions considered to be morally bad tend to be perceived as intentional, while the bringing about of side effects of actions considered morally good, or neutral, is generally taken to be unintentional. This appears to be particularly true in cases in which there is a moral judgment as well as a causal one: Knobe & Fraser (2008) show that moral judgments of an event may impact the causal judgment on said event. In the example above, say that David sings every night and wakes up his neighbours every time he does, and Mar is David’s mother. One could thus expect Mar to prevent him from singing at night. But we have seen in section 3.1 Mar does not care whether David sings or not, in other words she is indifferent to the outcome. Since the occurrence of the singing (and its side effect of waking everybody up) is borne out of indifference, people will be likely to judge that what she did was morally reprehensible, considering that she breaks rules of politeness and community. While not being entirely similar to McGrath’s normality, it appears to rely on similar mechanisms: if the speaker has any form of expectation (be it moral or not) with respect to the Causer’s behaviour, and those expectations are not met, she is more likely to judge that the Causer is causally responsible for the result.

Because conceptual indifference shapes how we frame and articulate causal relations, it must be recognised and integrated into existing frameworks. Therefore, indifference must be defined as a causal force that warrants modelling in its own right.

Intuitively, indifference can be understood as an attitude that is neutral towards a certain result. Put simply, the holder of said attitude has a tendency (in the form of an intention) that is neither for nor against the result. Let us take an example. Say that some people want to set up a small yard sale in their neighbourhood. The mayor could prevent them from doing it: he has authority over what is happening in his town. But it is a small town, and everyone knows everyone, so this should not pose any problem. The mayor will not get anything out of it himself, therefore, he has nothing to gain by allowing it nor by preventing it. In short, he simply does not care whether the yard sale takes place or not. By virtue of being a mayor, he has authority over (i) his constituents and (ii) whether the yard sale occurs or not. This is why his intention towards the yard sale matters: if he is in favour of it, it will take place. If he opposes it, it will not. If he is indifferent to it, the occurrence of the yard sale will depend on some other factor (e.g whether the inhabitants of the neighbourhood intend for it to occur or not). The indifference relation is summed up in Table 2 below:8

In short, the two bottom rows can be expressed through a laisser-causative such as the following one:

At first glance, it is unclear why a volitional entity who is indifferent (hence does not position themselves for or against a given result) would be fit to receive the role of Causer in a syntactic causative such as (25). Indeed, Causers of syntactic causatives are taken to have control over both causing and caused event. Therefore, we expect the Causer of a laisser-causative to influence the result, even when she is not directly responsible for it. What is highlighted by the data in Table 2, however, is that entities that are indifferent to the result fail to influence it. This mirrors the discussion in section 3.2 above: the (non-)intervention of the Causer seems to be neither necessary nor sufficient, or, in more trivial terms, “useless”.

What makes the mayor able to receive the role of Causer is his being the holder of preventing properties: he possesses authority over his constituents through his position as mayor. As such, he could prevent the yard sale from occurring, yet he does not. In sum, he acts as a would-be preventer for the occurrence of the yard sale. This example is not an isolated one, and the would-be prevention requirement is not forced by this specific scenario in which there is an authority asymmetry between the mayor and its constituents. In fact, Donazzan & Raffy & von Heusinger (2020); Raffy (2021); Donazzan et al. (2023) all point out that the French verb laisser in syntactic causatives requires for its Causer to be would-be preventer for the event encoded by the embedded verb. Any entity that does not have the relevant preventing properties is infelicitous as a subject of a laisser-causative. Obviously, the notion of would-be preventer can be applied more loosely to volitional entities, as they have free will and can choose to prevent events from occurring or interrupt ongoing events. Therefore, what makes the would-be prevention component salient is indifference: there is an expectation that those entities should act, or could act, to prevent the result, and yet they do not. In the following section, I show that indifference is crucial in order to (i) understand laisser-relations and (ii) define let as a causal notion that is distinct from enable: adding indifference to our causal modelling allows for a more fine-grained categorisation of “permissive verbs”.

Unlike the theories discussed above, causal models are not part of a particular approach to causation; rather, they are formal representations of causal relations. Causal models, in the framework developed by Judea Pearl (Pearl 2000; Pearl & Mackenzie 2018), consist of structural causal models (SCMs), which combine graphical and functional components to represent how variables influence each other. These models are typically represented as directed acyclic graphs (DAGs) depicting causal influences. The DAG structure is made up of nodes representing variables (each corresponding to a proposition) and edges (or arrows) that express direct causal dependence between variables. In this framework, variables are classified as exogenous or endogenous. Exogenous variables, which have no incoming arrows, are determined by background factors outside the model and are treated as independent. In contrast, the value of endogenous variables depends on the values of other variables: each such variable is associated with a function that determines its value based on those of its parent variables (i.e., the variables with arrows pointing to it, see Halpern & Pearl (2005)). This functional structure captures the underlying causal mechanisms and enables the model to evaluate how the system responds under different conditions. Thus, the values of the nodes are not static but can vary depending on the specific configuration of the model, allowing one to reason about interventions and counterfactual scenarios. Let us take a real-life example to make this clearer. When it is sunny outside, one generally puts their clothes out to dry them faster, because the warmth from the sun causes the moisture from the wash to evaporate. Therefore, if it is sunny (S), my clothes will get dry (C), hence S→C. This causal relation is represented in Figure 1.

In the model above, the node S is exogenous, since it is not influenced by another variable. Although it is clear that there are causes for sunny weather, these causes are not causally relevant to the situation depicted in the model. Conversely, node C is endogenous: its value is set by that of S, because whether it is sunny or not matters for whether my clothes will get dry. In other words, if S = 1, then C = 1, and if S = 0, then C = 0.

Now that the models have been clearer, and since they are perfect for representing causal concepts, let us apply them to a typical enable situation. Say that I want to represent the sentence in (26) below, which encodes a situation in which users would simply not be able to share their opinions on the platform if this was not an available feature of the website. In order to represent this situation accurately, we need at least three nodes: (i) whether the platform possesses an option to let users post comments, (ii) whether users intend to share their opinions, and (iii) whether users do share their opinions. What we are interested in here is how these nodes interact with each other.

The model in Figure 2 is a collider: the result node C acts as a collider in that it is influenced by at least two other nodes, here P and I_U. In this model, both P and I_U have a stimulatory influence, and both are necessary for the occurrence of the result. If the platform enables comments but users do not intend to comment, then they will not be commenting. Similarly, if users intend to comment but this is not an available option on the platform they are using, then they will not be able to share their opinions on that very platform. As we have seen above, this means that the value of C should depend on that of both P and I_U. This is summed up in the truth-table below in Table 10.

The data in Table 3 actually supports the observations previously made in section 3.1: in enable relations, the intervention of the Causer is necessary for the result. If P = 0, then C = 0, regardless of the value of I_U. This means that, if P is set to 1, it “wins” by setting the value of C to 1. This means that P is efficacious for C (Copley 2005; Copley & Harley 2015). Copley (2021) defines efficacy as follows: “efficacious(A,E) iff A → E is in the model, and the value of fAE→(A) \documentclass[10pt]{article} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \usepackage{pmc} \usepackage[Euler]{upgreek} \pagestyle{empty} \oddsidemargin -1.0in \begin{document} \[ f_\overrightarrow{AE}(A) \] \end{document} in the model is the value of E in the world”. Therefore, the notion of efficacy can be used to describe the participation of the Causer to the causal relation in enable situations.

Now that the functioning of causal models has been made clearer, and that the different types of influences have been addressed, these devices can be put to work to provide an accurate representation for let.

My approach to authorise is based around the analysis made for authorise-laisser in Donazzan et al. (2023), discussed above in section 2.2. As a brief reminder, the authors propose that the Causer in authorise-laisser has authority over the Causee’s set of alternatives; she could restrict that set but does not. I have argued above that the reason that she does not act on that set is because she is indifferent to the outcome. Nevertheless, what this entails is that the Causee has free choice over her alternatives: she is not made to pick one or the other and can act according to her own tendencies (desires, beliefs, intentions, abilities). Therefore, an additional criterion must be added in order to represent authorise-laisser: the Choice criterion.

We find that the model above in Figure 5 perfectly fits these requirements without any need for further modification (see Table 9): the Causee’s node, I_D, is completely exogenous, which means it is independent from any influence. Therefore, the Choice criterion is met. The authority of the Causer is represented by (i) the arrow coming out of P_J and pointing at the edge from P to S and (ii) the edge being made dashed by the Causer’s indifference (i.e. P_J being set at 1). The (amended) Prevention criterion is therefore met as well.

Let us turn to not-intervene. Recall that not-intervene refers to situations in which there is an ongoing event performed by the Causee that is not being interrupted by the Causer; in those cases, no alternative set is introduced, as the event is ongoing. This means that, while the question of free choice is not relevant for not-intervene (Donazzan et al. 2023), the issue here is that of the ongoingness of the caused event. Say that I utter the sentence in (34) with a not-intervene reading – the sentence can be paraphrased as (35), in which we find the periphrastic construction en train de, which encodes the progressive aspect.

The event encoded by the embedded predicate is already ongoing at t when the letting starts: they appear to be co-temporal rather than successive, unlike in authorise relations. In turn, this entails that the event cannot be prevented from happening but can be prevented from reaching its endpoint or culmination. Therefore, ongoingness must be added to the model.

Applying the model for let from Figure 5 above to our configuration in Figure 7 causes us to run into a problem with respect to the Ongoingness criterion. This problem arises because the most basic causal models we have been using do not represent time; therefore causal modelling cannot represent aspect. For that reason, they can simply never satisfy the Ongoingness criterion. The most obvious way to go about representing scenario like that of (34) and (35) is to actually represent it as in Figure 8 below, by treating the node as a constituent that is large enough to include aspect. In short, F is simply to be treated as F = whether my cousin is picking fruits.

This slight modification immediately repairs the model, which is now well-fitted for not-intervene.

Such an approach allows us to account for any type of situation involving a would-be preventer. While I have exclusively spoken of volitional entities in this article, they are not the only ones that can enter laisser-causatives as Causers. Non-volitional, non-force-generating Causers are perfectly felicitous (exclusively) when the Causee is also non-volitional and non-force-generating.9 Since the notion of indifference is restricted to agentive entities who can choose to act or not, it is obviously impossible to make it apply to constructions in which the Causer is not an Agent. The two questions that arise from this minor obstacle are the following: (i) do these constructions behave in the same way as their agentive counterparts, and (ii) if they do, how can we account for them without using the notion of indifference?

Let us consider the relation encoded by the sentence in (37) below, taken from Donazzan et al. (2020: 67).

The involvement of Causers of non-agentive laisser-relations is the same as that of agentive ones. The curtains are not necessary for the light to enter the room, a window with no curtain would have yielded the exact same result. They are also not sufficient for the light to enter the room, as the light would have failed to filter through if the window was tinted, for instance. Therefore, we observe the same “uselessness” of the Causer in non-agentive laisser-causatives.10 This fact leads us to expect that this apparent uselessness is actually a case of would-be prevention: the Causer possesses properties (that are not actualised) that are relevant for preventing the result. This leads us to our second question: what is encoded by the sentence in (37) is not that these curtains have the property to let light through. Instead, what is implied is that those curtains should not let light through, and yet they do. This is possible strictly because curtains are artefacts, and there are given expectations on artefact kinds. According to Thomasson (2003), artefacts, being man-made, come with built-in intentions. In other words, all objects of the same kind are expected to serve the same purpose since they were produced with the same intention in mind. In order to serve this purpose, they are created with a given set of properties relevant to their kind, which Thomasson calls ‘K-relevant properties’ (2003: 600). Curtains may typically serve two purposes: they may be decorative, or they can be used to block light. If the curtains are designed with the built-in intention for them to be decorative, then the only K-relevant properties that they may have are for them to be pretty or trendy (which is overall fairly subjective). On the other hand, if they are meant to block light, then the main property that they should possess is that of being opaque. It is clear that the curtains in (37) are not the decorative kind, as there is no sort of connection between being pretty and letting light through. Conversely, their opacity (or lack thereof) is what is at stake here: there is an expectation that they ought to block light, and yet they fail to do so.

What is being expressed here through the laisser-construction is a mismatch between an ideal world in which artefacts are perfect instances of the kind that they belong to and the real world in which artefacts can be defective. The expectation of the speaker with respect to the Causer is just not met; as such, Causers of non-agentive laisser-causatives are not would-be preventers, they are should-be preventers. They nevertheless fit the definition for normality in McGrath (2005), as one can consider that it would be normal for curtains to block light (since it is a K-relevant property of the kind curtain), and this specific curtain acts abnormally.

Therefore, I believe this model can be used to represent any situation of causation by omission. Let us conclude by revisiting our roast example from the introduction in (1), repeated below in (38):

I have treated let as encoding situations introducing a comparison between the actual world and an ideal world in which entities behaved according to the speaker’s expectations. What laisser does is underline the mismatch between these two worlds. This mirrors the notion of normality as defined by McGrath (2005): there are expectations for entities to behave a given way, which is what she calls normally. If they stray from such expectations, they become Causers because they are abnormal.

The function from A_f to the (non-)function from E_f to B represents this mismatch between the ideal world and the actual world. By making B independent from E_f, A_f sets it to be influenced solely by H. In other words, the only “real” reason why my food is burning is because heat is being applied to it even though it is already cooked. If heat was no longer applied to it (say, because my oven broke in the middle of cooking it), then my friend’s forgetting would not matter anymore because there would be no result.

Let us take a closer look at Table 10: the two rows in which B is receives no value (marked ×) are rows in which there is no expectation for on my friend to turn off the oven: a sentence like (38) simply cannot be true if such an expectation did not exist. The results for the two remaining rows are explained easily using the data for the influence relation from Table 8. When my friend does not forget to turn the oven off (A_f = 0), then the inhibitory function f– from E_f to B applies: my friend turns the oven off as expected and the roast does not burn. On the other hand, when my friend does forget to turn the oven off even though it would be normal for him to do so, then the independence relation R_indep applies; whether the burning of the roast occurs or not is now solely dependent on an outside factor (here, whether heat is applied to my roast or not). This means that causation by omission and let as a causal notion are actually one and the same.