lemÃŒc. Lemizh grammar and dictionary

A sketch of pragmatics I. Relevance

The fact is that human â€¦ languages do not encode the kind of information that humans are interested in communicating.

(Dan Sperber and Deirdre Wilson. Relevance)

What is clear from context?

Lemizh provides far-reaching possibilities not to express irrelevant information. We need not inflect words for person, number, or tense; there are no complements (required objects); and we can even omit certain case endings by compounding. This raises the question: what can really be omitted; what is ‘clear from context’? We will answer this question further down on this page, but we need to discuss some prerequisites first.

The linguistic field studying context is pragmatics. A significant portion of pragmatics seems to be universal or near-universal. So, our method will be to give sketches of the most important pragmatic concepts, followed by remarks on their implications for the Lemizh language.

Currently covered topics are

  1. Relevance: relevance theory, explicatures, real-time interpretation, and implicatures
  2. Triggers: scalar inferences, presuppositions, conventional ‘implicatures’, and non-restrictive constructions

Literature is given at the bottom of page II.

Relevance theory

Among the existing pragmatic frameworks, Sperber’s and Wilson’s relevance theory seems to be best suited for our purpose. It attempts to explain the wide range of pragmatic phenomena with a minimum of rules, making it attractive from a Lemizh viewpoint. The following account also draws on Carston’s work on this theory.

Let’s start with the definition of relevance, in the sense that relevance theory uses the word. Some clarifications will follow immediately.

An utterance – or any other observed phenomenon – is relevant to an individual to the extent that its positive cognitive effects on the individual are large and the mental processing effort to achieve these effects is small.

An individual’s ‘mental world’, i.e. the set of facts that are probably true from their viewpoint (everything they can perceive, remember or infer, including background knowledge that they are not currently aware of), is called their cognitive environment. Cognitive effects are effects on the cognitive environment, such as addition or removal of beliefs, increase or decrease of the confidence in existing beliefs, or the reorganisation of information into new patterns to simplify processing later on. Typically, there are more cognitive effects if the utterance contains new information that is somehow related to the addressee’s current cognitive environment, so that he can draw conclusions from the combined old and new data. Positive cognitive effects are those that are helpful rather than hindering for the individual, e.g. providing true information as opposed to false information. The mental processing is largely automatic and not accessible to the individual’s consciousness.

Relevance is not an absolute but a comparative property: the more positive cognitive effects and the less processing effort, the more relevant the utterance.

Here are some examples to illustrate the concept of relevance. If we’re planning to go on a trip next weekend and I tell you Next weekend the weather will be really awful. this is highly relevant to you, as you can draw a host of conclusions, such as: I want us to rethink our plans and want to inform you of this wish; you agree – or you don’t agree and just want to bring oilskins; I want to know your opinion on that matter; etc. By contrast, saying The weather was really awful on 19 October 1974 in Cumbria. gives you just one piece of new, completely unrelated information and is thus hardly relevant; and The weather is really awful right now. doesn’t tell you anything new, as you can see for yourself. Finally, On the weekend 2429 weeks after 19 October 1974 the weather will be really awful. contains exactly the same information as (1) but requires more effort to process, and is thus less relevant under our definition.

The principle of relevance

The communicative principle of relevance says that every utterance conveys the information that it is

  1. relevant enough for it to be worth the addressee’s effort to process it. (If the utterance contained too few positive cognitive effects for the addressee in relation to the processing effort needed to achieve these effects, he wouldn’t bother processing it, and the communicator needn’t have taken the trouble to utter it.)
  2. the most relevant one compatible with the communicator’s abilities and preferences. (Otherwise the communicator would have chosen a more relevant utterance – e.g. one that needs less processing effort and/or achieves more positive cognitive effects on part of the addressee – to convey her meaning. After all, she wants to be understood as easily and reliably as possible.)

We say that every utterance conveys a presumption of its own optimal relevance. If I tell you something – anything –, you are entitled to expect that I wanted my utterance to be consistent with the principle of relevance. Consequently, if I tell you something that does not seem to be worth your processing effort, such as (2) or (3) above, or something that seems to be less relevant than I could have put it, such as (4), you will automatically search for alternative interpretations. The most easily accessible interpretation that is consistent with the principle of relevance is the one you accept as the right one, because any further interpretations would cost you more processing effort and would thus violate condition (b). Your conclusion for utterance (4) might be that I want to test your math skills, or – more likely in these circumstances – that I want to illustrate a point about relevance theory.

The relevance-theoretic comprehension procedure
Follow a path of least effort in computing cognitive effects: test interpretive hypotheses in order of accessibility, and stop when your expectations of relevance are satisfied.

The constraint that utterances are compatible with the communicator’s abilities and preferences accounts for suboptimal communication, such as when the communicator is unable to think of a better phrasing at the moment, as well as for stylistic and cultural preferences (e.g. politeness considerations), withholding information, and lying.

The cognitive principle of relevance says that human cognition tends to be geared to the maximisation of relevance. It is the basis for the communicative pronciple of relevance, but we don’t really need it for our purposes.

We will now look at how the communicative principle of relevance guides us in finding out what information the communicator wanted to convey.



When I utter something, say Susan believes that her kiwis were too sour. you first decode it, but this only gives you incomplete information and normally does not even yield a full proposition (something that is unambiguously either true or false). Your job in interpreting it is far more complex: you need to infer a lot of additional material. The inference process is basically a search for the most relevant interpretation. It makes use of contextual information such as preceding communication, your environment and your real-world knowledge, and involves

We will have a closer look at most of these items shortly.

So, after the presumption of optimal relevance, the second piece of information communicated by my utterance is ↣ I have said that Susan Pevensie believes that the kiwifruit she, Susan, grew were too sour for the judges at the fruit grower’s contest. If I am talking literally (as opposed to, say, ironically), the utterance also communicates ↣ I believe that Susan Pevensie believes that the kiwifruit she grew were too sour for the judges at the fruit grower’s contest. If I have reason to believe that my utterance will convince you, it further communicates ↣ Susan Pevensie believes that the kiwifruit she grew were too sour for the judges at the fruit grower’s contest.

Such conclusions, which are developed from the utterance by referent assignment, disambiguation, construction of ad-hoc concepts and enrichment, are called explicatures; we will symbolise them with ↣. The proposition (8) is its basic explicature, while (6) and (7) are higher-level explicatures; (7), moreover, expresses my attitude towards the proposition (8). Naturally, this is called a propositional attitude I hold towards my utterance.

Propositional attitude

The utterance can have additional higher-level explicatures communicating my propositional attitudes. Some examples: ↣ I am bewildered that Susan Pevensie believes that the kiwifruit she grew were too sour for the judges. ↣ I feel sorry that Susan Pevensie believes that the kiwifruit she grew were too sour for the judges. or, if the statement is meant to be ironic, ↣ It is ridiculous to say that Susan Pevensie believes that the kiwifruit she grew were too sour for the judges.

For an utterance to be ironic it also has to echo someone else’s utterance or beliefs without saying so explicitly. There is not enough space on these pages to go into the details of echoic utterances, however.

I can give you various cues to indicate propositional attitudes, for example tone of voice, certain words or phrases such as ‘unfortunately’, ‘please’, ‘I am quite sure â€¦â€™, ‘I regret â€¦â€™, or grammatical features such as the mood of a sentence: the first explicature of an imperative utterance, i.e. its analogue of (6), is ‘It would be desirable that â€¦â€™. Depending on other cues and context, further higher-level explicatures of an imperative could be

Go north until you reach the river and then follow it downstream. ↣ I would be desirable to the addressee to go north until he reaches the river and then follow it downstream.

The first explicature of a question is ‘It would be relevant to know about â€¦â€™. In asking a regular question, the speaker expresses that the answer would be relevant to her; in rhetorical questions, she brings something to the addressee’s attention that she thinks is relevant to him. Neither imperatives nor questions have a basic explicature.

The future tense can be a cue that I am expressing an expectation, a hope, or else making a promise: The witch will never again threaten Narnia. ↣ I expect / hope / promise / assure you that the witch will never again threaten Narnia.

Often such cues do not unambiguously determine the propositional attitude, and you have to take the context into account (again).


All pragmatic inferences, i.e. all inferences that rely on context, are defeasible: they can be ‘defeated’ (cancelled) by explicit information or context without sounding self-contradictory. Thus I could cancel most of your conclusions from (5) by continuing I’m not talking about Susan Pevensie but about Susan Sarandon. She made the birds Chinese style, sweet-sour, but her guests didn’t like them. The mechanism of defeasibility is roughly as follows: an utterance is usually consistent with a number of sufficiently relevant basic explicatures, one of which is the most easily accessible and thus accepted in a given context. On encountering additional information that invalidates this explicature, the next accessible one that is still consistent with all available information takes its place. While this ‘switch’ of explicatures often feels unexpected or surprising, it gives the addressee no sense of contradiction as the new explicature is just as consistent with the original utterance as the old one.

Referent assignment

Assignment of referents is not only necessary for proper nouns but also for common nouns and noun phrases with a definite article when used referentially, such as ‘the witch’, ‘the green bridge downstream’, or ‘This ship’s captain is insane’ if we know who the captain is and specifically want to refer to him. On the other hand, nouns and noun phrases can also be used non-referentially or attributively, as in ‘This ship’s captain is insane’ when we want to refer to whoever the captain is. Generic reference is a reference to a whole class, as in ‘The beaver is a mammal’, and does not call for referent assignment.

Referent assignment notably includes deixis. Deixis refers to words, phrases and grammatical constructions whose meaning depends on the circumstances of the utterance, such as speaker, addressee, time, place, and accompanying gestures such as pointing. Examples are demonstrative pronouns, first and second person pronouns, tense, ‘here/there, now/then, yesterday/today/tomorrow, come/leave, Mum/Dad’, etc. Deictic words often take precedence over non-deictic ones, as they are easier to process: the present day is always referred to as ‘today’ and the next as ‘tomorrow’, sparing the addressee the effort to access the current date. As a consequence, saying ‘See you Saturday’ on a Friday or Saturday must refer to the next relevant Saturday after tomorrow, typically the following one.

Ad-hoc concepts and figures of speech

Illustration of concept narrowing (the ad-hoc concept is contained in the lexical meaning), concept widening (the lexical meaning is contained in the ad-hoc concept), and concept shift (ad-hoc concept and lexical meaning either overlap or don’t overlap)

An ad-hoc concept can differ from the lexical meaning of a word in several related ways: it can be narrower, as in the example above, or in Smilla has got a brain. ↣ Smilla has got a very well-working brain. Concept narrowing amounts to the addition of logical properties to the lexical meaning, in this case ‘very well-working’. An ad-hoc concept can also be wider, commonly referred to as loose talk and amounting to the removal of logical properties: France is a hexagon. ↣ France is an approximate hexagon. The photo booth mechanic is bald. ↣ The photo booth mechanic has very little hair. I see a rubber duck. ↣ I see a duck-shaped object made of rubber.

The combination of addition and removal of properties results in a shift of meaning. Thus, Turkish Delight comes in cubes of various colours. widens the meaing of ‘cubes’ to approximate cubes, as with hexagonal France, but also narrows it down by excluding cubes with edge lengths of, say, half a millimetre or half a light year. The shift can go so far that no overlap between lexical meaning and ad-hoc concept remains. This is how many figures of speech work: The witch exploded. ↣ The witch went into a violent rant. (metaphor) Cair Paravel is sending ships. ↣ The Narnian Kings and Queens are sending ships. (metonymy) You are the best of beavers! ↣ You are an extremely good beaver. (hyperbole)

For such an utterance to be understandable, the ad-hoc concept still needs to have some salient properties in common with the lexical meaning: structural properties in the case of metaphors, such as a sudden and unpleasent outburst of noise in (20), relational ones for metonymies, such as using the location for the government in (21), and qualitative ones for hyperboles (22).

Powerful metaphors (‘She is my anchor in the storm’) contain ad-hoc concepts with many slightly different meanings, and no specific one is exclusively communicated; so they achieve a wealth of expression that is not possible when speaking literally. Metonymies derive their strength from the ability to focus on a specific property of the explicated concept: ‘We need some good brains’ and ‘We need some good hands’ both refer to people, but to different qualities of them.

By the way, some utterances explicate a literal and a metaphorical meaning, as when a chess player tells his opponent ‘Your defence is an impregnable castle’. This causes problems for relevance theory as for other pragmatic frameworks.


Enrichment is similar to concept narrowing in that it adds logical properties, except that it does not apply to words but to grammatical structures and whole utterances. The difference is not always clear-cut. It is often useful to distinguish two kinds of enrichment, the completion of a semantically incomplete utterance to a proposition and the further expansion of an utterance that has been completed to, or already is, a proposition.


In the example (5), we have completed the genitive ‘her kiwis’ to ‘the kiwis she grew’ and the degree indicator ‘too sour’ to ‘too sour for the judges’. Other completions include ‘sour enough’ (for whom?), the comparative ‘more sour’ (than what?), the superlative ‘the most sour’ (of what?), as well as fixing the scope of negations â€¦ Many arrows didn’t hit the target. ↣ Not many arrows hit the target. OR Many arrows missed the target. … and of multiple quantifiers: All guests get a room. ↣ Each of the guests gets one room of their own. OR All guests get one room together.


An expansion turns a weaker proposition into a stronger (more specific) one, as in the following examples.

a weak statement (‘yellow flag’) with two disjunct strong interpretations (1: ‘completely yellow’; 2: ‘partially yellow’), the first of which is inferred
a weak statement (‘finger’) with two disjunct strong interpretations (1: ‘my finger’; 2: ‘someone else’s finger’), the first of which is inferred
a weak statement (‘finger’) with two disjunct strong interpretations (1: ‘my finger’; 2: ‘someone else’s finger’), the second of which is inferred

This flag is yellow. ↣ This flag is completely yellow. because, if the flag were only partially yellow, this sentence wouldn’t contain enough information for the addressee to picture the flag and would therefore violate condition (a).

Note the effect of different background knowledge: I broke a finger. ↣ I broke one of my fingers. I found a finger. ↣ I found someone else’s finger.

Here are two different expansions from the conjunction ‘and’: The bellman served out some grog and bade them sit down. ↣ The bellman served out some grog and then bade them sit down. The bellman told them about Boojums, and the baker fainted away. ↣ The bellman told them about Boojums, and therefore the baker fainted away.

Some further examples show that the literal meaning of an utterance can be quite uninformative, with only enrichment turning it into a useful piece of communication: ‘Repairing the ship will take time’ (how much?), ‘Dinner will get cold’ (how soon?), ‘I have had breakfast’ (how long ago?). As always, the answers to these questions are guided by the search for relevance: the time it takes to repair something is only worth mentioning if it takes more time than the addressee might expect; having had breakfast is only worth mentioning if I’m still full; and so on.

And here is a defeated enrichment: The Lemizh flag is yellow. It contains a green ellipse with an egret.

In Lemizh

The described process of explicature holds true for Lemizh, as probably for all human languages. We have already touched upon the fact that the literal meaning isn’t necessarily an explicature when we talked about the reality of the parole in unit 9 of the tutorial: there we said that the main predicate has grammatical reality, which is one degree more hypothetical than the real world.

Lemizh has no moods giving cues about propositional attitudes. Modal verbs and verbs of communication take their place; the imperative in particular can be expressed by a variety of such words, as can be exclamations.

Deixis is carried by demonstrative and relative pronouns, the latter only when pointing to the parole. The deictic function of tense, kinship terms, ‘come’, etc., also relies exclusively on pronouns, overtly or not. Referent assignment for relative pronouns is of course done purely by decoding. A demonstrative pronoun usually refers to something that cannot be targeted by a relative pronoun, because the latter are unambiguous and thus easier to process.

We can broadly distinguish five types of concept narrowing, including enrichment. The distinction between the two does not really make sense in Lemizh, so we will only speak of narrowing. It is licensed by Rule Six of sentence grammar, which allows the speaker to omit any information she thinks the addressee can supply from context, and conversely allows the addressee to re-introduce the missing information, i.e. narrow down the meaning.

Of course the ‘strong’ meanings can also be expressed explicitly if necessary, with the exception of type N.

O. Narrowing with objects

Filling in omitted pronouns is a typical case of O-narrowing, as are completions of the comparative, the superlative, ‘enough’ and ‘too’, and the translations of (25, 26, 27, 28). Another example is the ‘and’ construction, which literally says that the mentioned referents, among others, constitute the object in question, but implies that there are no other relevant referents besides the mentioned ones via a missing object ‘and nobody else worth mentioning’ or the like.

M. Narrowing with a compound modifier

N. Noninstantiability narrowing

R. Narrowing by rephrasing

S. Semantic narrowing

The verb dnà. ‘walk’ is understood as ‘at walking pace’ unless speed is irrelevant and we are merely talking about the mode of movement (because otherwise using nenà. ‘run’ would have more cognitive effects) and ‘with an aim’ unless aimedness is irrelevant (otherwise fràw. ‘amble, stroll’ would be the more relevant choice): the addressee S-narrows ‘walk’ with the negations of ‘run’ and ‘amble’. Likewise, ràt. ‘drive, steer’ is normally understood as ‘without mounting up’ (otherwise xàc. ‘ride’ would be more relevant).

Verbs of force such as làjg. ‘bend; break’ have a ‘deforming’ and a ‘destroying’ meaning. Either can be suggested by context or made explicit by using more specific verbs such as <z>à. ‘turn into a z shape’ versus skràp. ‘split, turn into parts’.

Metalinguistic negations of the kind ‘He doesn’t like her, he loves her’, where ‘love’ is a sub-category verb of ‘like’, become ‘He likes her and even (tmÌ.) loves her’ in Lemizh.

How many types of narrowing?

A tree showing the relationships between narrowing and widening types. Details see text.

Formally, M-narrowing is O-narrowing with a bracket that is subsequently inverted and turned into a compound. S-narrowing with a sub-category verb is equivalent to O-narrowing with this verb, which is then absorbed by its predicate (unless of course it is a negation). Furthermore, N-narrowing can be described as S-narrowing with a concept for which there is no word in the Lemizh lexicon. So rephrasing is the only type that cannot be naturally subsumed under O-narrowing.

Concept widening is licensed by Rule Seven of sentence grammar: just as saying ‘I see white mice’ allows for the white mice not to exist, saying (18) allows for a duck-shaped object be present instead of a real duck. There are two types of widening, qualitative and partitive.

Q. Widening with a qualitative case

You can and should omit the qualitative case suffix if the shape of an object does not relevantly deviate from a cube, such as will mostly be the case with Turkish Delight, and leave its addition to the addressee’s interpretation process. Conversely, explicit addition of the qualitative suffix implies that the deviation is relevant in some way.

P. Widening with a partitive case

As with qualitative widening, the addition of the partitive suffix should be left to the addressee unless the existence of other tools etc. is relevant. Saying ‘The wind is the cause for the door opening’ and meaning ‘The wind is part of the set that constitutes the cause for the door opening [the other parts, such as the fact that the door is ajar, not being relevant]’ is just as fine as saying ‘There are cubes of Turkish Delight’ and meaning ‘There are roughly cube-shaped bits of Turkish Delight [the deviation from the cube shape not being relevant]’.

In most situations, though, the presence or absence of partitives and qualitatives has a very relevant and well-defined function: if the speaker were to omit them at will, and the addressee were to add them at will, Rules Four and Five would lose their edge. Brackets would not equate their object with their predicate, and even simple sentences such as ‘Father Christmas gives Lucy a bottle’ would fall apart as we’d not be able to tell whether Father Christmas, Lucy and the bottle partake in the same action of giving.

Concept shift is of course a combination of narrowing and widening. Apart from figures of speech, tool nouns are an important application: on hearing jstù. ‘the means of sailing’, the addressee will P-widen it to ‘part [or all] of the set of means of sailing’ and then N-narrow it to ‘a sail’.

Real-time interpretation

The whole interpretation process runs simultaneously with hearing a sentence: when you hear the first constituent (‘Susan’), you immediately decode it and form hypotheses about the correct referent. You activate your knowledge about that referent, which may turn out to be useful background information for understanding the sentence. Then you ask yourself a question about how it might continue: in this case ‘What did Susan do?’. Equivalently, we can say that you form the hypothesis that the next constituent will be a verb.

On hearing the next constituent (‘believes’), you check your hypotheses and, if necessary, correct them. Again, you activate knowledge (frames, assumption schemas) associated with what you know about the sentence up till now. You form a new question: ‘What does Susan believe?’, equivalent to the hypothesis that the next constituent will be either a ‘that’-clause, or a noun phrase with the preposition ‘in’. And so on. So each constituent can act as a background to later constituents by prompting the right questions and thus preparing the ground for them, and/or as an answer to questions raised by previous constituents, adding relevance to the sentence.

In Lemizh – answers to a long-standing question

Word order

On hearing and decoding the main predicate of a sentence – say, dà. ‘give’ – you immediately form questions about its objects. Given the Lemizh plot, the most prominent question often is ‘Who is the sender?’ (nom). If enough context is available, you form a hypothesis: for example, if we are talking about your birthday, the sender might be me. This hypothesis can be confirmed or disconfirmed if the next word actually is a nominative, or it can be tentatively accepted if some other object follows. The next likely question is ‘What is the content?’ (acc); this can only be guessed at if we know each other quite well. Next, ‘Who is the recipient?’ (dat) practically answers itself: it is you. Thus, missing objects in the sense of Rule Six are tentatively assigned meaning. This is how O-narrowing works in Lemizh.

While there is no universally valid order of cases, questions for plot cases are typically followed by questions for causal, temporal and spatial cases, in that order; agent-centered cases follow action-centered cases; and qualitative cases follow their corresponding primary cases. A word whose accent indicates that a lower-level word is about to follow, especially if the referent is unclear, can prompt a question for a bracket, i.e. for its inner case. This is often true of predicates with a topicalised non-factive. (As hinted in unit 5, brackets are easier to process than coordinations, whose accent does not indicate that more information about this object is about to follow.) A predicate with an agentive accent of course triggers the question ‘Who is the agent?’, which overrides all others.

A different order of cases normally makes the sentence more costly to process, as some questions are only answered at a later point, and tentatively accepted hypotheses may be disconfirmed late in the sentence. The additional effort can be offset by other effects: most importantly, objects that aid in assigning the right referents, applying the right disambiguations, concept narrowings and widenings, and generally in activating the right background knowledge and forming the right hypotheses, should be placed before those objects for which this kind of information is needed. More reasons for deciding on an order of objects are given in the discussion of Rule Two in the tutorial. The ordering of a bracket’s predicate and object is governed by the same principles. In short: the sooner the addressee knows what the sentence is about, the better.

Typical hypotheses for persons include the speaker, the addressee, and other people that are present physically or within the conversation. Likely hypotheses for things also are those that are present physically or within the conversation, including ones associated with activated background knowledge – such as birthday presents in the context of birthdays.

Omitting objects

Now we can put our appeal to ‘clarity from context’ more precisely: if the communicator judges that the relevance principle will lead the addressee to the correct hypothesis about an object, based on what the communicator knows about the addressee’s cognitive environment, this object should be omitted. If mentioning this object does not produce any additional positive cognitive effects for the addressee, it has to be be omitted to reduce their effort. Conversely, including such an object will prompt the addressee to search for additional cognitive effects the communicator might have intended. — The example sentences in the tutorial, especially in the first few units, are much more verbose than in any realistic Lemizh text or conversation.

Qualitative cases

A qualitative object without the corresponding primary object leads to the hypothesis that the latter is fully characterised by not being identical to the former. (See the translations of ‘despite’ and ‘against’ and the last exercise in unit 13; the inner qualitative is based on the same concept.) This is why placing the primary after the qualitative object is perceived as misleading.

To mark a sentence as a comparison early on, we place the qualitative object(s) as far left as is practical, i.e. immediately after their primary counterparts.


Compounds are of course shorter than their uncompoundend counterparts, which saves processing effort; but determining the morpheme boundary and the lost case ending can mean additional effort. The compounds easiest to process are those with a commonly used modifier and lost case ending – such as the acc with negators, modal verbs, ‘male’ and ‘female’, and grammatical number, or the fact with ‘profession’ and ‘hobby’. Somewhat more costly are those with a common head, because in these the lost case has to supplied ad hoc by the addressee – nominal verbs, isms and philes, collective nouns and adjectives of predilection fall into this category. This is why compounds from brackets are formed with the more common, and usually less salient, component as the modifier, losing the case ending that is easier to reconstruct. (See the paragraph in the tutorial on compounding with ‘female’.) Compound words referring to familiar entities such as ‘redthroat’ and ‘teacup’ are also cheap.

Round numbers have two easily reconstructible case endings; the exponential number comes first because it is more relevant (as it tells the addressee the order of magnitude), and also because the epenthetic consecutive is analogous to the one in weighted adjectives and other similar constructions. Large exponential numbers are compounded such that the base, which is always the same anyway, comes after the exponent; but the exponent can’t go in first place for grammatical reasons.

All of these are easier to process than the uncompounded versions. Using the latter will lead the addressee to expect

Compounds with two uncommon components and non-obvious inner cases are often too costly to process to comply with the principle of relevance.

Relative pronouns

Recall this comment from unit 6 on the choice of relative pronouns. The cases of sibling objects are much more important for understanding a sentence than their order, and therefore more prominent in the addressee’s mind. This makes indirect reference usually easier to decode then direct reference. If the target word is the pronoun’s sibling and is immediately preceding it, direct reference with a type IIn pronoun is straightforward and usually to be preferered.


Given an utterance such as (5), if you do not find explicatures that meet your expectations of optimal relevance, you will activate more background knowledge that might be useful: ⇷ Susan is ambitious. If she loses at something, she’s pretty downcast. From (8) and (31) together you conclude ⇸ Susan surely is downcast right now. and, with further background knowledge, ⇸ Susan needs to be cheered up. ⇸ I want you to ring Susan and cheer her up. If I intended you to activate (31) in order to give optimal relevance to (5), it is called an implicated premise of (5) and symbolised with ⇷. If I intended you to infer (32–34) for the same reason, they are called implicated conclusions of this utterance, which we will symbolise with ⇸. Together these are the utterance’s implicatures.

There is a smooth transition to further ‘implicatures’ that might or might not have been intended by me: ?⇸ You should buy Susan some chocolate to cheer her up. ??⇸ You should introduce Susan to the Snark to cheer her up. We say such conclusions are only weakly implicated by (5), as I, the speaker, can hardly can be held responsible for them.

The distinction between explicatures and implicatures is not always clear-cut. Possible criteria are that implicatures are supposed to be linguistically and/or logically independent of explicatures, that negation of a sentence negates the explicatures but loses the implicatures, and that embedding the sentence in an ‘if’-clause preserves the explicatures but not the implicatures. None of these criteria is foolproof, however.

Here is another example: if I have been to the opera and tell you The bellman produced a series of sounds corresponding closely to the score of an aria from The Hunting of the Snark. this costs you more effort to process than if I had just said The bellman sang an aria from The Hunting of the Snark. You presume that I used the most relevant phrasing and conclude that I intended additional positive cognitive effects to offset the effort, such as that the bellman’s performance was so bad it couldn’t really be described as singing, and that I wanted to express a humorous or cynical propositional attitude towards his abilities.

Implicatures are pragmatic inferences and are therefore defeasible without self-contradiction: But don’t ring her up. She wants to be alone right now. provided this doesn’t ruin the relevance of the utterance. Contrast this continuation of (37): But I’m not being cynical. which only adds to the cynicism.

The following classical example illustrates the context sensitivity of implicatures. Planning their vaccation in France, Peter and Susan discuss visiting their old acquaintance Amélie. Peter: Where does Amélie live?
Susan: Somewhere in the north of France.
If it is clear that Susan wants to visit Amélie, this implicates ⇸ Susan doesn’t know where exactly Amélie lives. On the other hand, if Susan clearly knows her address, it implicates ⇸ Susan doesn’t want to say where exactly Amélie lives. ⇸ Susan doesn’t want to visit Amélie.

In Lemizh

The mechanisms for deriving implicatures in Lemizh are the same as in any other language.


We now see that the ‘literal meaning’ of an utterance can be quite irrelevant or even false and the utterance still meaningful by the explicatures and implicatures it yields. The utterance is just a piece of evidence from which, together with contextual information, the intended meaning has to be inferred.

But why does language work that way; why doesn’t it just consist of logical propositions, as was traditionally thought? The probable explanation is that nonverbal communication of feelings, intentions and wishes developed well before verbal communication, so that the mental apparatus for inferring information from relatively weak evidence was already available, and a completely (or just largely) explicit way of verbal communication would have been redundant.

Lemizh vs. English

In Lemizh, we have to use pragmatic inference more often than in English, because its grammar provides more possibilities for omitting information. But Lemizh also uses literal expressions in many situations where English relies on pragmatics, such as: