I have been considering what differences or similarities in any properties at all could be found between:

  • a language where whatever the supposed “deep structure” of a language truly is (like, the ultimate logical/semantic form, I guess) is fully explicit in strings of a language - therefore, it is deterministically parsible (and this means, it is guaranteed to obtain a parse: there is never incompleteness or insufficient information to parse totally - and only one parse: there is never ambiguity or multiple possible parses.)
  • The opposite, in some way. This means, however the language is used, there is a degree of non-determinism in it. That could mean that there are approximation techniques which people use to guess what deep structure element some surface structure element has (like resolving a pronoun to its referent), and they may be effective in 90% of cases, but 10% of the time, it is necessary to ask for clarification.

The question can be contextualized by looking at some similar, interesting phenomena.

  1. I read that sign languages may have anaphora mechanisms that are quite different from spoken language, like “tethering” a referent to a particular area or object in the room.
  2. I do not know any programming language that permits implicit anaphora, but it’s an interesting question if it could be useful. That means, the computer should statistically guess what the referent of an expression, based on how recent it was.
  3. A non-deterministic language can be more extreme than just having a few statistically resolved elements in the parse. Perhaps it is an inherently stochastic or aleatoric language, in which the use has something to do with the language’s ability to generate an unexpected meaning, each time the same sentence is said (or anything more out of the box like that, for its conceptual value).
  4. Possibly, the factors at play include: a. How many possible parses are there? b. What is the probability for each one? c. What elements of sentence creation and parsing have a relationship to those questions?

Maybe it is a simple formula, but I would like to see it developed. Maybe we can consider a single global metric for the amount of steps required to produce a message in its final form, and to parse it.

Then, it may be something like this:

-> let us abstractly imagine S is the set of all possible statements in the language of thought (LoT). We choose to ignore their internal structure or relationship to one another and simply list each with a unique identifier, a, b, c, d, etc.
-> For some reason, there is a transformation system that symbolically changes the LoT into a “transport form”, for it to arrive at the recipient of the message. The recipient is able to parse the incoming message and restore it to its original form, thereby obtaining the true message.

Consider the transformations/mappings/rules/programs necessary to transform a message into transport form, and the reverse. Mathematically, it may be the case that they are not true inverses of one another.

I think there might be a formula between the size of the generator program, the size of the message, the size of the parser program, and the mathematical goal of the communication system (to transport messages with perfect one-to-one fidelity, or not, and why?). For the last one, we should decide a criteria on the system to be fulfilled, and it could be many different things. Whatever criteria are selected becomes an optimization metric on the system. We could have a system that was selected to optimize for message brevity. It’s an open question, but maybe this would imply that the generator is trivial: it maps one element from LoT to one element in a transport medium. There is a lot of thinking to be done, but the point is, you can notice how different systems achieve different things, and not claim all human languages share an identical optimization explanation.

It reminds me of something you hear nowadays that all languages are equivalent in their expressive capability, but it’s not clear what argument was given to demonstrate that. I recently felt that Chinese is easier to use on a smartphone because you type a few letters, and the keyboard suggests various characters. I’m not sure, but I wonder if you use your fingers less in Chinese, on smartphones.


We might ask:

Assuming we have some good mathematical data on situations of ambiguity, do we know, say, the average accuracy, speed, and even brain energy consumption, with which, say, English speakers resolve ambiguities, in language? (I now realize that ambiguities goes far behind anaphora, but even includes homophones or similar-sounding words.)


1 Answer 1


The claim that "all languages are equivalent in their expressive capability" is true yet doesn't mean what you think it means. It only means that every proposition can be somehow expressed in every human language. Compare this to bee language, which can express direction, distance and quality of a food source, but cannot express the proposition that there is no food at a certain location (or that the "speaker" is hungry).

Not every language uses exactly the same means to express a given proposition, indeed depending on how you calculate "means", no two languages do. For instance in Logoori, there is a verb root "cheen" which expresses the idea that one reacts to a misdeed directed at one, where the reaction is supposed to cause a sympathetic reaction from a higher power, especially ancestoral spirits. We don't have a single word for that in English. There have been many speculations about the comparative amount of energy required to utter some string, such as [papi] versus [pabæ]. Nobody knows how many calories are required for an individual to articulate those strings, or for an individual to compute the derivation of [pabæ] from /mpəpʔa/, not to compute the meaning "I used to cook it". My finding is that English is hard to type on a cellphone, because the keyboard is so tiny. I have no idea what average user-experiences are for fluent cellphone user who know Chinese, vs. in English. Differences may have to do with autocorrect technology.

Anaphora exists in all human languages. There is always a way to say "I found him", and there is no language which requires you to uniquely identify the referents (e.g. "Citizen 243843969 found Citizen 5986595856"). The possibility of one-to-many mappings between utterance and proposition is a certainty in human languages. In no language is the semantic interpretation fully explicit in the realized utterance. Languages do differ in the extent to which some of the vagueness might be grammatically regulated. In English, we just say "Bill saw Tom", but in Logoori you have to commit to a more specific time (and other stuff) – such as "a week ago; two days ago; within the past two days; recently within the day; just now"), and actually there is no generic "some time in the fast" form that is compatible with "3 years ago" and "just now".

  • Thank you. Well, why do you think human languages are all less than 100% “formal”? The general answer is always “economy”, right? But that needs some sort of model to make precise. If I were to assert, “a language would be much better if there was zero anaphoric resolution, let’s make a language with a one-to-one mapping between “propositions” and “utterances” and actually use it”… could the answer be, “Sure, do it - the state of language is arbitrary - emoji were designed by humans, modern Hebrew is a constructed language. People can learn new languages.” ? Commented Jun 8, 2023 at 18:40
  • Sort of like looking at mountains and asking, “why do they always slope?” - maybe, there’s not any deep mystery, it’s just happenstance given particular factors which caused them to exist (or, “why do stalactites always point downwards?”). But there might be fascinating patterns nonetheless. Why are pronouns so common, but not pro-adjectives, for example? I think we need a model with reproducible generative character, for how language evolved, to feel confident why language is how it is and not some other way. The fact that humans can intervene in their own language is a significant Commented Jun 8, 2023 at 18:44
  • “confounding factor”. Commented Jun 8, 2023 at 18:44
  • Anyway, I think a cool thought experiment is to try to design new variants of English. One does not allow any explicit linking of elements. People are forced to try to guess what the relationship between some simultaneously presented elements is. Even a sign language with “no grammar”, has the ordering principle of time. On the other end, to explore a variation of English with features to eliminate anaphora. Maybe with use, one can decide what about such languages they do not like. Maybe an obvious reason anaphora is centered around nouns is that nouns are entities, which in a way appear Commented Jun 8, 2023 at 18:51
  • like the most fundamental unit of language, since clauses tend to be subject-predicate, universally (I think). Commented Jun 8, 2023 at 18:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.