A sketch of pragmatics II. Triggers

Do I claim anything by asking whether the mouse over there, who is talking to Lucy about elephants, is the valiant Reepicheep?

While the phenomena discussed on this page are also often context dependent to some degree, they are mainly tied to certain words and grammatical structures. We will therefore talk of sentences rather than of utterances carrying them. This is merely for convenience; the exact mechanisms are – as so often in pragmatics – disputed.

Scalar inferences

What has traditionally been called ‘scalar implicatures’ are inferences triggered by words that can be arranged in scales from weak to strong, such as ⟨some, many, all⟩, ⟨or, and⟩, and the definite number terms ⟨one, two, three, …⟩. They have been variously argued to actually be implicatures or cases of enrichment and thus explicatures. For our purposes, this makes no real difference.

The witch turned many Narnians into statues. While this is logically compatible with The witch turned all Narnians into statues. it is usually inferred that → The witch didn’t turn all Narnians into statues. This inference arises because, had the witch actually turned all Narnians into statues, saying (2) would have been more relevant, as the addressee could have drawn additional conclusions from the complete absence of unharmed Narnians. Therefore, (1) wouldn’t meet condition (b) of the relevance principle. In general, scalar inference means that on encountering a weaker statement we conclude that the next stronger claim on the scale is not met.

Propositions (1) and (3) taken together give the final interpretation: ⇒ The witch turned many but not all Narnians into statues.

Depending on context, the inference might also be → I don’t know whether the witch turned all Narnians into statues.

Here are some more examples: The witch turned some Narnians into statues. → The witch didn’t turn many Narnians into statues. ⇒ The witch turned some but not many Narnians into statues. We’ll be needing a Snark or a Boojum. → We won’t be needing both a Snark and a Boojum. ⇒ We’ll be needing either a Snark or a Boojum. If I find the password I’ll be able to read Jacopo’s texts. → If I don’t find the password I won’t be able to read Jacopo’s texts. ⇒ If and only if I find the password I’ll be able to read Jacopo’s texts.

Being pragmatic inferences, scalar inferences are defeasible: Did the witch turn many Narnians into statues? — Yes, even all of them. We’ll be needing a Snark or a Boojum, and if possible both.

Numerals and other non-entailing scales

In contrast to the mentioned scales, where the stronger statement entails (logically requires) the weaker one, stand non-entailing scales, the prototype of which is the scale of definite number terms. Others are ⟨calm, breeze, gale, storm, hurricane⟩ and, in the context of getting actors’ autographs, ⟨Emma Croft, Bob Peck, Richard Harris, Julia Ormond⟩. Compare the following examples with (9): Has Smilla got four chairs? — No, she’s even got six. Did you get Bob Peck’s autograph? — No, I even got Julia Ormond’s.

There are however complications. In some situations, a number term seems to entail the possibility of lower or higher numbers: You may borrow four books. → You may borrow fewer than four books. because you may or may not borrow each of the four books.

You must read four books. → You may read more than four books. because from ‘You must do A (= read four books)’ can’t possibly follow ‘You mustn’t do B (= read one or more additional books)’. Analogously, Peter: I need four more chairs for my party.
Susan: Smilla has got four chairs. → Smilla may have got additional chairs, i.e. more than four. but Peter: How many chairs has Smilla got?
Susan: Smilla has got four chairs. → Smilla has got exactly four chairs. because Peter’s question implies the exact number would be relevant to him.

The correct analysis of number terms is still debated; but the described view is good enough for our purposes.

In Lemizh

In Lemizh, not only definite, but also weighting numerals constitute non-entailing scales. So, ‘The witch turned many (dmÌ.) Narnians into statues’ literally says that she didn’t turn all of them into statues. Likewise, (9) becomes ‘Did the witch turn many Narnians into statues? — No, all of them.’ Contrast this with the indefinite numeral in ‘Did the witch turn several (mlÌ.) Narnians into statues? — Yes, even all of them.’

This also means that we have to translate ‘many, perhaps all Narnians’ as naRniÌ dmynÌn jnÌnyn. Narnian-acc¹ 3/4-partacc-partacc² 1/1-partacc-partacc². ‘many or all Narnians’, and ‘Not many arrows hit the target’ as ‘Few arrows hit the target’ (see unit 8, Modified objects).

In other words, most scalar inferences aren’t inferences, or a matter of pragmatics, in Lemizh; they are purely semantic. The ‘or’ and ‘if’ examples, however, constitute narrowing by rephrasing. (See exclusive ‘or’ and Conditional clauses in the tutorial.)

Presuppositions

A presupposition of a sentence (symbol: ») is an uncontroversial background assumption that is necessary for that sentence to achieve relevance. For example, ‘Edmund regrets eating the sweets’ can only be relevant if he actually ate them – at least under normal circumstances – so the presupposition is ‘Edmund ate the sweets’. If the presupposition is not met, the sentence is trivially false and thus violates condition (a) of the relevance principle. Similarly, ‘Cratylos made fun of Hermogenes’ can only be relevant if Cratylos and Hermogenes existed; otherwise the sentence has no referents, likewise violating condition (a).

Presuppositions survive negation of their sentence, modifying it with a modal verb or adverb, and turning it into a question or an imperative: if he didn’t eat the sweets, ‘Edmund doesn’t regret eating the sweets’ is trivially true; ‘He must regret eating the sweets’ is trivially false; ‘Does Edmund regret eating the sweets?’ answers itself; and ‘Edmund, regret eating the sweets!’ cannot be obeyed. None of these are worth any processing effort. We will use the catch-all term modified sentences for all of these, for reasons explained below.

Here is an overview of words and grammatical constructions that trigger presuppositions. This and the following two classifications (conventional ‘implicatures’ and non-restrictive constructions) are mine; you will find other analyses in the literature. The links lead to examples in the tutorial.

* As mentioned, questions in general share the presuppositions of the corresponding statements. However, wh-questions are also presupposition triggers.

† Presuppositions that are prerequisites for a sentence to have referents are called existential presuppositions. Apart from definite descriptions (P9), these are triggered by temporal clauses and phrases as well as connectives such as ‘again’: if he had never been sitting there, (P10) refers to a nonexistent time span, and so does (P11) if I hadn’t met her before.

Properties

Positive sentences entail their presuppositions: if Edmund regrets eating the sweets (P1), it normally follows that he must have eaten them. If she stopped singing (P4), it logically follows that she had been doing so before. If Cratylos made fun of Hermogenes (P9), we can conclude that both of them existed. As entailments, they are non-defeasible. Interestingly, we can construct situations in which presuppositions of positive sentences can be defeated: He regrets having eaten the beaver. He hasn’t yet realized that it had only been a dream. I saw the white mice. Most people see them when they are completely drunk. Presuppositions of negative sentences and questions are not entailments but purely pragmatic inferences, and are therefore defeasible (or don’t arise in the first place): She didn’t close the door. It has been shut since this morning. Did she manage to dance? — No, she didn’t even try. He hasn’t eaten the sweets, so obviously he doesn’t regret eating them. Sentences modified with modal verbs or adverbs fall in either of these categories (‘must, should’ etc. vs. ‘might; probably’ etc.). The imperative is usually not defeasible.

In contrast to implicatures, presuppositions can sometimes, but not always, be detached (removed) by omitting their trigger: She managed / didn’t manage to dance. → She danced / didn’t dance. I didn’t see the white mice. → I didn’t see (any) white mice. Is it her who is building bridges? → Is she building bridges? Who has eaten the sweets? → Has somebody eaten the sweets?

The presupposition may be part of the addressee’s background knowledge already before the sentence is uttered, or it may be new information that the addressee can effortlessly add to their cognitive environment. The latter situation is called presupposition accommodation and is much more common than implicature accommodation: on hearing ‘She didn’t stop singing’ (¬P4), the addressee will usually just accept that she had been singing. If you didn’t know I had a dance partner, you will automatically accommodate to both the existential and the change-of-state presupposition in Sorry I’m late. My dance partner didn’t stop talking. By contrast, (¬P11) cannot be easily accommodated; an addressee who knows nothing about any previous meetings will inquire or at least silently wonder about them.

Presuppositions frequently – again, not always – survive in situations that are more complex than the aforementioned modified sentences. This is the notoriously unsolved problem of presupposition projection. Compare: If Edmund regrets eating the sweets, everybody will be happy. » Edmund has eaten the sweets. (projects, i.e. survives) If the sweets contained sorbitol as a sweetener, Edmund will regret eating them. » Edmund has eaten the sweets. (projects) If Edmund ate the sweets, he will regret eating them. » ∅ (doesn’t project) Lucy thinks / claims / is telling everyone that Edmund regrets having eaten the sweets. » ? (no consensus among people) The witch thinks / claims / is telling everyone that she is the queen of Narnia. » ∅ (existence of a queen of Narnia doesn’t project, even for people who don’t know the background story) The children aren’t telling anyone that they are kings and queens of Narnia. » There are kings and queens of Narnia. (projects, even for people who don’t know the background story)

There is currently no theory which correctly predicts in all situations whether a presupposition projects.

Finally, presuppositions can be misused to present information as an uncontroversial background assumption that in fact isn’t. The first example is a harmless case of ‘implicit’ communication, the other two are rather sneaky: Amélie is really cute. — Yes, and her boyfriend is very attractive too. » Her boyfriend exists, i.e. Amélie has a boyfriend. Stop being passive-aggressive! » You are being passive-aggressive. What will you give me for my birthday? » You will give me a birthday present.

In Lemizh

Much of what has been said about defeasibility, detachability, accommodation, projection and misuse is (nearly) language independent. Detaching existential presuppositions arising from definite articles (23) isn’t possible in Lemizh as there are no articles – on the other hand, there is a simple and powerful way to make presuppositions non-defeasible, to which we will get shortly.

Pragmatic reality (P1, P8, P9, P10)

As in English and other languages, most positive and some modified sentences entail their presuppositions; and many others imply their presuppositions via the principle of relevance. In terms of Lemizh grammar, all of these are cases of pragmatic reality, i.e. reality of objects that is not licensed by Rule Seven of sentence grammar but by logical or pragmatic inferences. Here are the translations of (P1) and (¬P1), which both presuppose that he ate the sweets.

Those presuppositions that are not derived by entailment but via relevance are of course defeasible.

‘Edmund doesn’t regret eating the sweets’ would of course be translated with a modified object for the sweets.

Grammatical reality (P1, P8, P9, P10; P2, P6, P7)

Sometimes, though, it is desirable to confer grammatical reality on a presupposition using a bracket with the main predicate (which amounts to a topicalised predicate) or an inversion, depending on which word is more important for understanding the sentence in a given context. This is useful whenever there is uncertainty about a defeasible presupposition, as can happen with negated sentences such as (¬P1) and sometimes with positive sentences.

Such constructions also translate judgemental ‘that’-clauses (P2) and English clefts (P6, P7), as well as their negations. This gives us a natural way to detach these presuppositions: we only have to undo the topicalisation or inversion, as we did with judgemental clauses in unit 15.

These constructions ‘demote’ the original statements to pragmatic reality: the first example does not grammatically claim the action of not regretting it but the content, i.e. the eating. Technically, the object that is pragmatically real is the missing factive object – just as the giver in ‘Lucy gets a bottle’ is pragmatically real via the missing nominative object. We will call this kind of non-obvious pragmatic reality cross-case reality. The main point here is that Rule Seven can only make one claim per sentence grammatically real; nonetheless, it is rarely necessary to split original statement and presupposition into two sentences.

‘It isn’t good that you speak’, ‘It isn’t her who’s building bridges’, the inverted phrasing of ‘He doesn’t regret eating the sweets’, and analogous phrasings of other modified sentences are actually modified objects; hence the terminology.

‘know’ is translated as ‘have learned, have heard, have seen’ or the like, none of which necessarily presupposes the truth of the accusative object; therefore, we need grammatical reality if there is any doubt about the truth of the ‘known’ thing. On the issue that ‘see’ doesn’t necessarily presuppose its object see also the example ‘I didn’t see Amélie watching TV’ in unit 14.

Implicative verbs (P3)

For implicative verbs, only the presupposed action (‘try’) has its own verb, and the presupposing action (‘manage’) is expressed with the same verb, using grammatical reality: ‘manage’ is either the topicalisation of the tried action, or the inversion of ‘try’. The presupposition is only pragmatically real in either case; there is no easy phrasing with grammatical reality, and it’s usually entailed anyway.

Negating or otherwise modifying ‘manage’ can be done with cross-case or grammatical reality for ‘try’. The cross-case construction negates an accusative bracket of two verbs – a construction we have already met in unit 13 when we talked about counterfactual statements. To make ‘try’ grammatically real, we simply invert.

Changes of state (P4)

ganìR. ‘She stops singing’ entails the action of singing via cross-case reality; grammatical reality is neither easy to phrase, nor is it necessary. The presupposition of the negated topic can be made grammatically real by uncompounding and inversion.

Implicit changes of state (P5) and comparisons (P11, P12)

There is no easy way to either grammatical or pragmatic reality for these presuppositions; you have to express them explicitly if you need to. (Lemizh expresses ‘again’ as ‘like earlier’, so it is grouped together with comparisons here.) However, as with the other types, they are mostly either entailed or pragmatically implied.

Also recall that ‘The Neapolitan isn’t as good a rider as the Frenchman’ has to be translated as ‘The Neapolitan is a less good rider than the Frenchman’.

Conventional ‘implicatures’

This is the traditional name for information encoded in a sentence that does not contribute to its basic explicature but adds something extra. This is done in an implicit way, via certain triggers – in contrast to the constructions discussed in the next chapter. For example, ‘Achilles is fast but clumsy’ explicates that Achilles is fast and clumsy, and additionally conveys a contrast between speed and clumsiness via the trigger ‘but’. Each of these two pieces of information has to satisfy the relevance principle independently: if the conventional ‘implicature’ is not met, the sentence sounds inappropriate, the explicature can still be relevant.

The actual mechanism of conventional ‘implicatures’ is a matter of ongoing dispute, but they are certainly not implicatures in our modern sense. To avoid this name, we will just call them CIs, which can also be read as ‘coded implicitly’. We symbolise them with yet another arrow: ⯮.

Like presuppositions, they survive negation and other modifications, but for a different reason: they fall outside the scope of these operations. — The folowing links mostly lead to dictionary entries.

* (P3/C3) presupposes that she tried to dance (because she couldn’t manage if she hadn’t tried), but has the CI that dancing is difficult for her (because this piece of information is independent of her actually dancing). The verbs ‘remember’ and ‘forget’ convey presuppositions when followed by ‘that’-clauses but CIs with a quite different content when followed by infinitives. There are many more implicative verbs triggering a large variety of CIs.

Implicative verbs also typically imply that the content of the subordinate clause actually happened, or else that it didn’t happen. These are either entailments (‘She managed / remembered / was forced / didn’t hesitate to dance’ ⊨ She danced; ‘She didn’t manage / forgot / was not able / refused to dance’ ⊨ She didn’t dance) or implicatures (‘She wasn’t forced / hesitated / had intended to dance’ ⇸ She didn’t dance; ‘She was able / didn’t refuse / hadn’t intended to dance’ ⇸ She danced).

† We have already talked about connectives in the presupposition chapter. If they refer to something nonexistent, this usually renders a sentence irrelevant as in (P11) – but sometimes relevance can be saved: the word ‘yesterday’ in (C4) already estabilshes a time span, and so we don’t have to rely on ‘again’ having a referent. This seems to blur the distinction between presuppositions and CIs.

The connective ‘also’ illustrates nicely that CIs have to satisfy the relevance principle: when I say ‘I also met Amélie’, the other person(s) I met must be relevant in the context.

The referent of the connective can also be mentioned explicitly: from (C5) we infer that the ancient object found before was the chess knight (as there is no other candidate that satisfies the relevance principle), and so the chess knight must be an ancient object.

‡ The comparison (C7) is distinguished from (P12) by the fact that, should Nerissa not be a girl, we are still informed that Portia is taller than Nerissa; while (P12) becomes utterly irrelevant if the Frenchman isn’t a rider. Again, there is only a thin line between presuppositions and CIs.

Properties

Being pieces of encoded information, CIs are not defeasible. They can however be consistently detached by changing the trigger that encodes them to a neutral phrasing, or just by omitting it: Achilles is fast and clumsy. The tortoise is shouting at Achilles although he fed it. She danced. I met her yesterday. They discovered a chess knight. Afterwards they found some ancient objects. I will be able to read Jacopo’s texts if I find the password. Portia is taller than Nerissa.

Accommodation and misuse aren’t a thing for CIs as they are not background information.

CIs in subordinate clauses fall outside their scope and so survive practically all kinds of embedding. While this could be legitimately called ‘CI projection’, I have never encountered the term. Probably the only exception is the analogue to (29), which gives us the nonsensical **If I have a negative attitude towards the tortoise, then the damn tortoise is shouting at Achilles. Embedding a CI in a clause expressing indirect speech also projects. This has been called speaker-orientation: if I say ‘The tortoise thinks that Achilles is fast but clumsy’, I subscribe to the view that there is a contrast between speed and clumsiness. Also, I wouldn’t say ‘Achilles told me that the damn tortoise was shouting at him although he had fed it’ if I didn’t at least share Achilles’s negative attitude towards the tortoise. Apparent exceptions such as ‘The tortoise told me that even he was once overtaken by Achilles’ probably qualify as irony.

In Lemizh

As mentioned above, Lemizh can only make one claim per sentence. This typically means that a CI translates into a pragmatically real part of the basic explicature (unless of course you split the sentence in two); sometimes, however, the CI becomes grammatically real in the Lemizh translation and the original basic explicature is demoted to pragmatic reality. Both ways, CIs fall inside the scope of negations and other modifications – including subordinate clauses –, and are therefore often defeasible. For the same reason, they do not necessarily project. Detaching mostly works simply by omitting the object or compound modifier that carries the meaning, except where noted below.

Pragmatic adverbs (C1) and expressives (C2)

The pragmatic adverbs ‘but, even, already, finally’ translate as tmà. ‘lead to the opposite expectation [given previous information]’; ‘even’ needs an inner partitive; ‘already, finally’ are temporal objects. See the verbs of certainty for how this verb’s plot works; ‘assume’ is a near-antonym, though tmà. specifically relates to information given in the discourse at hand. Pragmatic ‘therefore’ is the opposition of ‘but’. Expressives are verbs of emotion with an inner dative/accusative (‘the disliked’) or causative/persuasive (‘the reason for dislike’). Honorifics are of little importance in Lemizh.

Negating the first example literally gives ‘It is not true that Achilles is fast and unexpectedly clumsy’. This leaves the decision what part(s) of this claim are false entirely to pragmatic inference, i.e. the unexpectedness is defeasible. The same is true for the other examples.

Implicative verbs (C3)

The translation of ‘manage’ was treated under the presupposition trigger (P2) above. ‘try’ implies that the tried action is difficult in Lemizh as it does in English, and can be made grammatically real if desired. Detaching works as usual for the positive sentence – by simply omitting ‘try’ –, but the negation needs some juggling with inversions to arrive at ‘She didn’t dance’. If the difficulty is the main point, we can express it directly.

‘avoid’ and ‘abstain’ are exactly parallel to ‘don’t manage’, but with the modal verbs ‘must/should’ and ‘want’, respectively, instead of ‘try’. Their CIs, too, can be detached by removing the modals and leaving the negator in place.

‘happen to do’ translates to a factive bracket meaning ‘do by chance’; ‘remember, forget, confess’ translate straightforwardly as verbs. In short, there is no common pattern in the translation of implicative verbs.

Explicit connectives (C5)

The Lemizh solution is to include the connected property (‘ancient’) in the first sentence and reference it from the second one with a pronoun in a qualitative case, circumventing identity of action. Naming the property in the second sentence gives a meaning similar to the detached version (C5′).

By the way, the bracket translating ‘chess knight’ having ‘chess piece’ as predicate and ‘horse’ as object is a good example of placing objects first that aid in assigning the right referents and applying the right disambiguations.

The other tiggers (C4, C6, C7)

The other triggers are unsurprising: they are translated as shown in the tutorial and the dictionary (linked from the table above) and modified in the usual way. Detaching the CI of (C6) needs a factual conditional, i.e. a switch of the main predicate from negation to uncertainty.

Speaker-orientation

Speaker-orientation is not built into the Lemizh equivalents of CIs because they fall inside the scope of indirect speech, as mentioned. We therefore need to express it explicitly if necessary.

In this example, the PI_n−5 pronoun cannot be part of what Achilles told me because it points to my parole. In general, speaker-orientation can be achieved by weak linking with a contextual case. (The linked example, ‘You said that you would visit your uncle, a wise man’, is actually a non-restrictive construction, to which we will turn now.)

Non-restrictive constructions

A non-restrictive construction encodes information that, like a CI, does not contribute to the basic explicature. It is inserted into a sentence to give rise to an additional basic explicature, but in an explicit way. Like CIs, non-restrictive constructions fall outside the scope of modifications and subordinate clauses. Actually, they are sometimes classified as CIs, but we will treat them separately.

The two basic explicatures in each of these sentences are processed separately by the addressee. This can be seen by comparing the explanatory and circumstantial clauses with either two separate sentences (which also have two basic explicatures) or a compound sentence with ‘and’ (which only has one): The man forgot to buy the artichokes; he was speaking to the child about elephants. The man forgot to buy the artichokes and he was speaking to the child about elephants. In structures with two basic explicatures, one can be interpreted as an explanation for the other; thus, (37) is typically expanded to ‘The man forgot to buy the artichokes because he was speaking to the child about elephants’ just as the explanatory and circumstantial phrasings. This interpretation isn’t possible for the compound sentence (38).

Properties

Being encoded like CIs, non-restrictive constructions are not defeasible but detachable, and they are speaker-oriented: if I say ‘Susan complained that the poodle, which is in fact extremely well-behaved, is yelping’, I am the one claiming that the poodle is well-behaved; and if I say ‘Edmund believes that the dunce talking to the beaver is a butcher’, I am calling that person a dunce.

In Lemizh

As discussed in unit 16, translating a non-restrictive relative clause as a bracket is appropriate whenever the referent can be clearly identified beforehand. These translations have only one basic explicature as opposed to two in the English sentences. In other words, the bracket is only pragmatically real and consequently can be defeasible, especially in modified sentences. Also, brackets are not speaker-oriented; and they are detachable simply by omission.

The difference between English ‘two-explicature’ constructions and Lemizh brackets is nicely illustrated by questions. If the poodle is actually yelping but wasn’t born at midnight, the question Is the poodle, which was born at midnight, yelping? would be answered in English with Yes, but it wasn’t born at midnight. where the affirmative refers to the main clause. By contrast, in Lemizh we would answer No, but it is yelping. where the denial refers to the complete claim that (39) asked for, which is basically the logical conjunction of ‘The poodle is yelping’ and ‘The poodle was born at midnight’.

The alternative translation is of course as a separate sentence or a weak linking. In modified sentences, this moves the clause out of the modifier verb’s scope, making it non-defeasible; in indirect speech, it becomes speaker-oriented. Detachability remains unscathed.

Strictly speaking, there is no grammatical rule saying that the first sentence has to uniquely identify the referent. However, if the speaker were to move part of the identifying information away from the word or phrase in question (here ‘poodle’, ‘man’, ‘Nechwatal’, ‘Reepicheep’, and ‘[the one] talking to the beaver’), this would mean unnecessary processing effort for the addressee; so he normally understands it as additional information, i.e. as non-restrictive.

Other attributes (N3, N4) behave exactly the same way in Lemizh, being translated as brackets, separate sentences, or weakly linked.

Explanatory (N1) and circumstantial clauses (N2) are translated into Lemizh as conjunctional clauses, typically in causal cases. Along with referentially used descriptions (N5), they are pragmatically real, just like brackets: they are potentially defeasible, not speaker-oriented, and detachable by omission.

Conclusion II

Taking everything into account – especially entailment of scales, number of explicatures, scope of modifications, defeasibility of CIs and non-restrictive constructions, and speaker-orientation –, we conclude that translations from English into Lemizh (and vice versa) often have quite different pragmatic implications.

A sketch of pragmatics III