Friday, 15 March 2013

carrying dogs

English intonation is at the interface of phonetics and pragmatics. To describe the tunes you have to be able to analyse the changing pitch of the voice and the associated stress patterns, which is phonetics. To describe their meanings you have to be able to account for language use and contextual meaning, which is pragmatics. I feel confident about the phonetics, less so about the pragmatics.

Yesterday as I came up the escalator at Vauxhall tube station the man in front of me was carrying a medium-sized dog. He was complying with the instruction displayed at the foot of every escalator on the London Underground.

Eighteen months ago Language Log carried an interesting posting by Mark Liberman on this topic — in fact an expansion of one he posted as long ago as 20 Aug 2006

I've never figured out a really convincing explanation for why stressing "dogs" seems to encourage the interpretation "everyone must carry a dog", while stressing "carried" encourages the interpretation "if you have a dog, you must carry it".

Nor have I. I think this puzzle was first pointed out to me by Michael Halliday in about 1964; he didn’t have a satisfactory explanation, either.

If spoken, to carry the intended message this sentence must have the nuclear accent (“phrasal stress” for Liberman) on the verb:

  • (ˈ)Dogs must be ˈcarried.

If you say it with the nucleus on Dogs, you encourage the interpretation “you can't use this facility unless you are carrying a dog", rightly characterized by Liberman as absurd:

  • ˈDogs must be (ₒ)carried.

But why?

Please, pragmatics people, do go ahead and expatiate on the ‘implicit universally quantified agent’ by everyone and on the deontic modal must, because I don’t know how to, or at least not how to tie them up with the presence/absence of sentence accents.

In section 2.21 of my English Intonation, the section entitled Topic and Comment in the Tone chapter, I wrote

The topic is typically said with a non-falling tone (a dependent fall-rise or rise), the comment with a falling tone (a definitive fall).

OK, dogs is the topic, must be carried the comment; and we get the correct interpretation if we say

  • \/Dogs | must be \carried.
  • /Dogs | must be \carried.

— but this doesn’t explain why

  • \Dogs must be ₒcarried.
pushes us towards the absurd interpretation.


  • \Safety boots must be ₒworn.
or the injunction below — where the corresponding interpretation is not absurd at all, but the intended one.


  1. "I've never figured out a really convincing explanation for why stressing 'dogs' seems to encourage the interpretation 'everyone must carry a dog'".
    This seems bizarre to me. If I wanted to produce this meaning, I would stress "must" rather than "dogs". Stressing "dogs" sounds really weird.

    1. "Stressing "dogs" sounds really weird."
      That's the point. The question is, why?

      If you put the nucleus on "must", that would throw contrastive focus onto the deontic element, implying that someone is contesting the obligatoriness of the injunction. But it doesn't resolve the ambiguity: there are still the two possible readings.

    2. Indefinite noun phrases can be semantically ambiguous in their reference.

      The most amusing example is the ambiguity of She wants to marry a Norwegian. In one interpretation there's a specific Norwegian who hasn't been identified. In the other, there is at present no Norwegian available, but she lives in hopes.

      There's some similarity with Dogs must be carried. No specific dogs are being referred to. They are hypothetical; there's a disguised conditional to the effect

      If there are dogs.

      I half remember from many years ago a device that I half understood then. It involved symbolic logic and propositions such as It is true that X exists such that X is a dog. Informally, then, since I can't remember the formal wording or notation,

      If anything which is a dog exists, then it must be carried.

      At this point, pragmatics kicks in and leads us to understand that the rule applies to any dog that is currently under the control of somebody using the escalator, and that this user must carry his or her dog while on the escalor.

      So Dogs is the condition and must be carried is the order. The latter carries what David Brazil calls proclaiming tone (generally main stress and falling intonation). The former carries referring tone (=not main stress, not falling intonation). It's no different from

      If you knows a better hole, go to it.

      By contrast, there is nothing hypothetical about Safety boots.

      [OK, that's a pragmatic judgement. Semantically, the sentence could amount to If you have a pair of safety boots, you must put them on and wear them. Our knowledge of the world includes the observation that people on building sites wear their boots. They don't go round barefoot holding them away from the mud, for example.]

      The sentence means Everybody must wear safety boots implying that they must wear them at any point beyond the notice, or else go away.

      So now it's must be worn that carries the referring tone. What else would you do with safety boots under the circumstances? and Safety boots is everything that calls for proclaiming tone: topic, focus and new information.

      Orders and decorations will be worn is different in various ways, the most important being:

      Although will may express an order, we can tell that it's no more than an expectation.

      Somebody in the Army can order a subordinate to wear decorations at a formal event, but it almost always goes without saying — and would never require the visual aids of the example above.

      [Actually, the Army avoid will. The order would be Orders and decorations are to be worn.]

      If we rephrase it as a request, then we could use the Referring-Proclaiming pattern:

      If you have any orders or decorations, please wear them

      But the notice isn't even a request. It deliberately chooses a modal will that is ambiguous; it could signal an order, but equally could signal a prediction

      Some people will be wearing orders and decorations

      leaving the reader to possibly do the same.

      My preferred intonation would play down the hypothetical nature of the orders and decorations and acknowledge the real-world knowledge that they're like boots. Apart from leaving them alone, what you do with them is wear them.

      So, for me, will be worn invites referring tone and orders and decorations invites proclaiming tone (main stress, falling intonation). The same pattern as Safety boots will be worn although semantically and pragmatically somewhat different.

  2. Doesn't it simply answer the question \what must be carried? This leaves out the step why \something has to be carried in the first place, but the change from the regular topic/comment pattern implies that.

    1. By "it" I assume you mean the version with the nucleus on "dogs". On the other hand the version with the nucleus on "carried" answers the question Dogs must be \whatted?, what must we do with dogs?
      I can't see that this gets us any further.

    2. Doesn't "what must we do with dogs?" assume we have dogs already - now what shall we do about them?

      My first thought was that voice (diathesis) and so, word order, do play a role, prompting a default assignment of topic and comment, even if intonation can overrule those. But the safety boots example shows that this isn't so simple.

      At any rate, having the nucleus on must wouldn't make it all logical and unambiguous at all; it would still leave the question open whether this means one must in general or in case you come with a dog already.

      And I don't think English is unique in this, or even Germanic languages, though other languages might use other means to differentiate, such as different expressions in the first place, or determination.

      I find this issue less problematic than to explain to an American or an EFL learner why we say apple 'pie and the like.

      (Phillip Minden)

    3. This American finds "apple 'pie" as American as apple pie.

    4. Perhaps he was thinking of "apple sauce", which notoriously has late stress in BrE, early stress in AmE.
      But let's stick to the point, please, people.

  3. I feel like I'm missing something here, as this seems a straightforward effect of topic vs. focus. Topics restate old information: dogs can only be the topic if there are contextually relevant dogs, which in this context means if you, the reader, are carrying a dog or dogs. The dogless reader has no such context to apply. Putting emphasis on dogs makes it the focus rather than the topic, and the focus represents new information, so that version of the sign is telling us to ensure we have a dog in order to comply with it.

    I think the reason it doesn't much matter in the case of the "safety boots" example is that all readers are presumed to have safety boots, so they are in any case old information. I can't understand the "orders and decorations" example; I presume it says to wear orders and decorations if you have them, but signs with will in them are new to me except for plain futurity, like Bridge will be closed Sunday.

    If I were writing the sign, I'd write Carry your dog, by analogy to the various dog-related signs here in New York like Curb your dog and Clean up after your dog, which obviously do not apply to the dogless. But perhaps the plain imperative is too rude for BrE?

    1. This is exactly what I was thinking: topic vs. focus, new information vs. old news. I can't believe that no pragmaticist has described it along these terms, or fathom why it apparently eludes at least Mark Liberman and John.

      As for the safety boots, it's difficult to imagine one could do something else with them besides wearing (carrying them?) so that makes it less of a problem.

  4. I see three reasons why the absurd proposition "Everyone must carry a dog" may correspond to this English sentence.

    1. If "must be carried" is unaccented, then the competent listener will assume that something must be carried, the only question is what. Actually it is forced on them that this is something already known.
    The same applies if a falling nucleus is used on "dogs", and a rising one on "carried". (Perhaps this latter intonation suggests it even more strongly that everyone must carry a dog.)

    However, this in itself leaves open another interpretation, namely that it is dogs, rather than something else that must be carried. But why do we think that there ARE, or MUST BE any dogs at all?

    2. The solution lies in a pecularity of English syntax. This can be illustrated if we translate the whole thing into Hungarian. (There is no equivalent of "carry" in Hungarian, I shall use the very colloquial word "cipel" for sake of simplicity).

    a) A kutyákat CIPELNI kell. = The dogs-t tocarry hastobe. This corresponds to "(The) dogs must be CARRIED".

    b) A KUTYÁKAT kell cipelni. = The DOGS-t hastobe tocarry. This means "It is (the) DOGS that must be carried.".

    c) ?KUTYÁKAT kell cipelni. = DOGS-t hastobe tocarry. This means that there are dogs that must be carried.

    It is impossible to have a written Hungarian sentence with the same ambiguity as in English, because i) The word order is different, and ii) the definite article cannot be left out if we are not talking about any general dog that may not be here.
    I don't know how ungrammatical "?The dogs must be carried" would be, but it certainly wouldn't be ambiguous. A fortunate wording could be "Any dogs must be carried".

    However, the Hungarian sentence c) still doesn't really imply that you have to have a dog OF YOUR OWN. It would rather mean that there will be dogs at the top of the escalator that will have to be carried somewhere. (With a strange emphasis that they are indeed dogs. The natural intonation of c) would be Kutyákat kell CIPELNI.)

    So where does this implication came from?

    3. It results from a semantic-pragmatic property of the word "carry". This passive form of this verb is very often used to mean that every person is performing the action.

    This utterence would similarly imply that you must have a dog with you:

    DOGS must be brought/Kutyákat kell hozni (intonation irrelevant in H.)

    But I don't think this one would to most listeners:

    DOGS must be lifted up/Kutyákat kell felemelni

    And this one certainly doesn't:

    DOGS must be muzzled/Kutyákra kell szájkosarat tenni

    1. Levente

      I don't know how ungrammatical "?The dogs must be carried" would be, but it certainly wouldn't be ambiguous.

      It would be grammatical but meaningless in this context.

      In another context it would mean 'You see those dogs? They must be carried' or 'Those dogs we were talking about must be carried' or 'Our dogs must be carried'. They must be specific dogs whose identity is known to both speaker and hearer.

      the definite article cannot be left out if we are not talking about any general dog that may not be here

      If you mean what I think you mean, that's not so in English. I saw her with a dog means 'I saw her with a specific dog which most certainly was there'. Hence the ambiguity of She wants to marry a Norwegian. 'I can see dogs' means 'I can see some actual specific dogs, and I can see them because they are here'. However, I would like to own a dog and He's afraid of dogs refer to an unspecified dog or general dogs, none of which are here.

      To make matters worse, English can express 'any general dog that may not be here' inthree ways when it's the subject of a certain type of generalisation.

      1. The dog is a domesticated wolf
      2. A dog is a domesticated wolf
      3. Dogs are domesticated wolf

      Use [1] with definite article is severely restricted. It can only refer to each and every dog that has ever existed or might exist. It cannot refer to each and every dog that happens to be near a particular escalator.

      By the way, is your sentence [a] ambiguous? Can it mean both 'All dogs must be carried' and 'The dogs in question must be carried' ?

      I think the simple fact is that Hungarian and English have different ways of expressing generic reference.

      I wonder if you observe the same difference as we do between the dog-carrying and order/decoration-wearing examples.

    2. Dear David,

      "?The dogs must be carried" [...]

      It would be grammatical but meaningless in this context."

      What I meant by ungrammatical is that it wouldn't be a well-formed sentence according to the rules of E. as opposed to those of H. for the same meaning as that of "Dogs must be CARRIED".

      "the definite article cannot be left out if we are not talking about any general dog that may not be here"

      You completely misunderstood me here, let me explain.

      Wat I said in the quotation doesn't refer to English, but to HUNGARIAN. What I was trying to say is that it is impossible to write an ambiguous H. sentence like "Dogs must be carried" for two reasons: 1. In "A kutyákat cipelni kell", the nucleus automatically goes to "cipelni", and it is ungrammatical to put it on "kutyákat". If you place the nucleus on "kutyákat", the word order changes to sentence b. 2. In a version which requires everyone to have a dog, like "DOGS must be brought"/"Kutyákat kell hozni", a definite article MUST NOT be used, however, in the H. version of E. "Dogs must be CARRIED", there MUST be a definite article.*(note 1)

      "'I can see dogs' means 'I can see some actual specific dogs, and I can see them because they are here'."

      This is the same in Hungarian, which means that I was wrong when I said that the zero article cannot be used when the dogs are "here".
      "I can see dogs" in H. is "Kutyákat látok" (Dogs-t isee). Furthermore, "Selling dogs is allowed" can be (d) "Kutyákat szabad árulni" (Dogs-t allowed tosell) - zero article, same as E., generic; but "Caressing children is allowed" is (b) "A gyerekeket szabad simogatni" (The childs-t allowed tocaress) - definite article, different from E., but also generic.
      (a) is much more common as "Szabad kutyákat árulni" or "Kutyákat árulni szabad". With definite article, "A kutyákat szabad árulni", it would be about specific dogs.
      (b) with zero article is rather strange with neutral tonic placement, otherwise it expresses contrast.

      "He's afraid of dogs" must have the definite article in H, except as a possible reply to "He's afraid of cats".

      "Use [1] with definite article is severely restricted. It can only refer to each and every dog that has ever existed or might exist. It cannot refer to each and every dog that happens to be near a particular escalator."

      In H., [1] is the only possibility. However, the H. equivalent of "the dog" also cannot refer to dogs near the escalator!*(note 2) The only possible expression is "the dogs" ("a kutyák"), in the plural! "The dog" (a kutya) can only refer to one specific dog.*(note 2)

    3. "By the way, is your sentence [a] ambiguous? Can it mean both 'All dogs must be carried' and 'The dogs in question must be carried' ?"


      "I think the simple fact is that Hungarian and English have different ways of expressing generic reference."

      What I realize is that there are many forms of expressing general reference, and the use of articles is incredibly complicated, I suspect, in both languages, but certainly in Hungarian. Now after thinking about it a lot, I highly doubt that pragmatics is the main conditioning factor. Lexical and grammatical nuances seem to me to be much more important.

      "I wonder if you observe the same difference as we do between the dog-carrying and order/decoration-wearing examples."

      No. "If you have a dog, please carry it" = "A kutyákat cipelni kell". Likewise, "Those who have decorations will wear them" = "A kitüntetéseket viselni fogjuk", not "*Kitüntentéseket fogunk viselni", which would mean that everybody will wear decorations. And my English competence doesn't detect any difference between "DOGS must be brought." and "SAFETY BOOTS must be worn.". For me, both indicate that you can't go there without dogs/safety boots.

      But to go back to my original analysis, the only reason I brought up Hungarian was to illustrate that an ambiguity arises in English because you can dislocate the tonic without losing an article or altering the word order, which is not true in all languages.

      Note 1. In a very informal register of some lects, the definite article can be omitted.

      Note 2. It is possible that in some non-standard lects, singular "a kutya" may be used for general reference. I would never use this structure.

    4. Levente

      an ambiguity arises in English because you can dislocate the tonic without losing an article or altering the word order

      I think it's much more general than that. Indefinite article a(n) and zero article before plural nouns are always prone to ambiguity, whatever the syntactic context.

      By definition, they never carry definite reference, but in many contexts they can carry either specific or generic reference.

      Now, there can be contexts where reference is a mixture of generic and definite. Let's stick with escalators. Many now have coloured margins at the side of each step. To warn passengers against the danger of getting long clothing caught in the gap, the advice is

      Stand between the yellow lines (or whatever the colour is).

      Note that it can't be Stand between yellow lines Although the warning refers generally to any step, it also refers to the specific and identifiable lines on any given step. And you generally stand on just one step. Even if your feet are on two steps, your weight is on one foot. Grammatically, it makes no difference; if you think you're standing on two steps, the warning refers to four specific and identifiable lines.

      English differs from some languages in insisting that the definite article is used in generic contexts only when the referend is unique and/or identified in each instance of such a context. So we can speak of the individual and the society but only when it's a single example. Otherwise *the society is just plain wrong. We can't say *The society was simpler in the Stone Age.

      The use of the in The dog is a domesticated wold is an anomaly in English. It doen't mess up the system, because the dog can't have its usual meaning in generic sentences like this. Even so, we only use it when the logical structure is:

      X is [differentiating characteristic] Y where Y is a the class which X belongs to

      The dog is a domesticated wolf
      The safety boot is an important precaution on building sites

      Another peculiarity of English is that generic the is compatible only with singular noun forms. It's grammatical, if unidiomatic, to say A dog must be carried (generally, at all times) but neither grammatical nor idiomatic, nor even intelligible to say The dogs must be carried at all times.

  5. This comment has been removed by the author.

  6. What about implicit contrast? I mean, we could say that by accenting "dogs" we are contrasting it with "other things (which must be carried)" or the need for such a dog with its lack, in the case of someone who's carrying nothing.

    1. Mariano

      Yes, I was planning to post to that effect.

      DOGS must be carried is what one might shout to somebody on an escalator with their arms full of shopping bags and their dog seated at their feet.

      In contests where there is no alternative object in view that is being or might be carried, then DOGS must be carried means 'You can't use the escalator unless you're carrying a dog'.

    2. David, Mariano, Levente, isn't all of this ust what I wrote above?

    3. Lipman :)

      Yes, it's funny, everybody is immersed in their own thoughts. I found Mariano's question surprising, because we had already discussed that issue.

      But what I think is the solution to the mystery isn't just about topic/comment, intonation, and word order. Somehow point 3) of my first comment is missed, although that is the key point of my analysis.

      Of course, the question still remains: why is it that "DOGS must be brought", "DOGS must by carried", and, I think, "SAFETY BOOTS must be worn" imply that everybody must have a dog/safety boot, but "DOGS must be lifted up" and "SAFETY BOOTS must be cleaned" do not?

      Professor Wells was right in saying that this is pure pragmatics (perhaps not even a linguistic question?).

  7. I think you're ALL missing the point. Please re-read what I originally wrote. Then discuss how we might design, for a text-to-speech system designed to read notices aloud (e.g. for the blind), an intonation component that would correctly place an intonation nucleus (presumably, a high-fall accent) on “carried” in “Dogs must be carried” but on “Safety” in “safety boots must be worn”. I want an ALGORITHM that could in principle be implemented in software.
    Correct nucleus location is in my view the most important problem not yet entirely satisfactorily solved in text-to-speech systems.

    1. Trying to clarify the problem, I dreamt up a scenario with a text-to-speech app in a blind person's smartphone.

      The immediate question is: Why does the blind person need more information than a sighted person reading the notice?

      Accepting that necessity, why not place the information in the notice rather then in the app? Put a chip in each notice that will alert a passing blind person via his/her smartphone. The user can then choose to have the spoken version played by the smartphone, complete with intonational information

      But, OK, if you must have the nucleus placing within the software of the app, then you'll need to introduce some rudimentary world-knowledge.
      • Include a number of common notice schemata that can be recognized, including X must be Y-ed
      • Store the information that what you do with dogs is not usually to carry them
      • Store the information that what you do with safety boots is usually to wear them
      • Instruct the program to stress X for ' yes, usually' and Y-ed for 'No, not usually'.

      This will generate
      Dogs must be ↘CARRIED
      ↘SAFETY boots must be worn.
      However, unless you're careful it will also generate
      Medals and ↘DECORATIONS will be worn, which may not be the intended interpretation.

      So, you need to be able to correct the app, and to input new world knowledge — to teach it new notice texts. This may not be feasible in the immediate future, but in an ideal future word the app should be able to interrogate the user.

      On recognizing the X must be Y-ed schema and identifying X as dogs and Y as carried, the app asks aloud:
      * What to people usually do with dogs? Do they usually carry them?
      When the user answers No, the app stores the information for future readings.
      *What do people usually do with safety boots. Do they usually wear them?
      When the user answers Yes, there had better be a supplementary
      Do people often do something different with safety boots?
      The answer No is stored and X is stressed.

      Similar interactive routines should teach the app new values for X and Y in the schema and norms — usual or unusual —for each X is Y-ed possibility. Ideally, the app should be able to learn new schemata, but that would be extremely ambitious.

    2. David: Very well, but now multiply that by all the ambiguous words there are. In the end, you wind up storing the intonation of every possible sentence, which is impossible. What would such an algorithm even ask the user if confronted with "British left waffles on Falklands" or "Club fight blocks rail river tube plan", or "Ike beats Tina to death"? Such a program isn't just ambitious, it's inconceivable without full human intelligence and a huge amount of background knowledge, and even then, most anglophones are going to have trouble with "Kiwi dole bludge rort probe quashed".

      There is an art to writing safety-critical messages in such a way that they cannot be misunderstood even for an instant. "Dogs must be carried" simply fails that test, and should be replaced forthwith.

    3. Joh Cowan

      In the end, you wind up storing the intonation of every possible sentence,

      You don't, because this hypothetical app is restricted to reading notices And not even all notices, but notices that conform to particular schemata. To start with, it might have just the one X must be Y-ed stored.

      Another constraint is that the program doesn't store new X's o Y-ed's until after interrogating the user.

      My only ambition for the program would be that it could recognise a lot of common notices and assign one pragmatic rule.

      And, as I said, I would think it more practicable to place the spoken version of the message on a readable chip within the physical notice.

      "Dogs must be carried" simply fails that test, and should be replaced forthwith.

      I don't believe it needs to pass the test. It isn't designed to catch the passenger approaching the escalator and urge them to pick up any dog before it's too late. It's more of a reminder Don't forget to pick your dog up addressed to people already on the escalator. OK, some people may not be previously aware of the requirement, but it's not the end of the world if they take a second or so to absorb the message.

    4. John Cowan wrote : "There is an art to writing safety-critical messages in such a way that they cannot be misunderstood even for an instant. "Dogs must be carried" simply fails that test, and should be replaced forthwith."

      I respectfully disagree. Anyone with a dog will understand the notice immediately (assuming that he or she speaks English), and thus the dog's safety will be safeguarded; anyone not carrying a dog may smile for a moment at the possible ambiguity, but none will feel obliged to snatch a dog from a nearby dog carrier in order that he or she (the potential dog-snatcher) may comply with the letter of the law. Thus the notice achieves all that it sets out to achieve and does not need to be re-worded or replaced.

      Philip Taylor

    5. John, you want an algorithm that understands pragmatics? You like aiming high, don't you? :)

  8. Dear Professor Wells,

    I see, that's an entirely different question. I thought you were interested to solve the "puzzle" that Liberman posted about. If the topic is where we put the nucleus to express the intended meaning, then your post was extremely cryptic.

    However, I'm not sure I even understand this question. In linguistics, there are no such things as "Dogs must be carried" and "safety boots must be worn", to which some intonation could be assigned. There are only utterances, which already have one intonation. If we represent an utterence by written signs, which do not indicate intonation, data is irrecoverably lost. In the signs in question, even the meaning becomes ambiguous.

    Even humans can only guess what the intended meaning is, based on their knowledge about the world, and then they can recover the utterance for that meaning (proposition).

    But a computer not only does not have any knowledge about any world, it has no access to any meaning! Even if it had the information that safety boots are usually worn, it doesn't understand the meaning of either "safety boots" or "worn". It sees something like "safetyN bootN plural mustAUX beAUX wornVpp".

    For it, intonations suggesting the propositions "Safety is that boots must be worn", "You must have your safety boots on", "Certainly, what people usually do with safety boots is wear them", "What you must do with your safety boots is wear them" or "There are safety boots here that somebody has to wear", among others, are equally correct.

    The only workaround this is to build a computer that has AI.

    1. "The only workaround this is to build a computer that has AI." Yes, this is a problem for pragmatics.
      Speech-to-text systems are important, and in daily use in e.g. airports. The operator types in an announcement (which could be an entirely new one, not a previously recorded one), and the speech-to-text system speaks it aloud. Kurzweil systems do the same thing reading books etc for the blind.

    2. Yes, I find the approach legitimate. It's between linguistics in the narrower sense and the prediction of a situation. I suppose the technical TTS answer would include the choice of grammatical voice as well as probabilities tied to the lexeme, if possible in the given context.

      It would fail often enough, as there are situations that are ambiguous to non-artificial intelligence, too. You could as well ask for software to know if the answer to something is yes or no. (Which isn't far-fetched, but AI rather than linguistics.)

      But why stressing _dogs_ implies you're not getting in without dogs doesn't seem so enigmatic, in other words if you know the parameters, you can predict a fitting intonation.

  9. This comment has been removed by the author.

  10. I observe that :
    1) (ˈ)Dogs must be ˈcarried. written French is :
    Les chiens doivent être portés. (with a definite article ("Les"), which in French is playing the part of a GENERIC article).

    (or, in oral French :
    Les chiens, on doit les porter).

    2) ˈDogs must be (ₒ)carried. the French language is :
    Des chiens doivent être portés.
    with an indefinite article. Just as in English,
    the sentence sounds ill-formed. A more natural version of it would be :
    2) On doit porter des chiens.
    I call this a more natural version, since the normal use of the indefinite article is to accompany a nominal group receving the end-focus.
    "Des chiens" signals something new, not given.

    So my opinion is that [ 2) ˈDogs must be (ₒ)carried. ] is an ill-formed version of "One must carry dogs". In both versions, the new information is "DOGS", so it does not have the intonational properties of a topic but those of a comment.

    Jérôme Poirrier
    Grenoble, France