Loading ...
Sorry, an error occurred while loading the content.
 

WSL: Weird-syntax-lang revisited

Expand Messages
  • Logan Kearsley
    So, the acronym is begging for this to be a sign language. That s something I ve wanted to do for a while, but I don t think I want to tackle it all at once
    Message 1 of 9 , Jul 27 5:44 PM
      So, the acronym is begging for this to be a sign language. That's
      something I've wanted to do for a while, but I don't think I want to
      tackle it all at once right now. Clearly, I will need a better name,
      but for now, it's WSL.

      The whole "universal language" and what're-the-basic-rules-of-language
      discussion has got me inspired to take my little weird-syntax-lang
      sketch and flesh it out into something more complete and useful. The
      basic idea is to, starting with the sketch I've already got, look at
      all of the basic compositional semantic operations in some model of
      formal semantics and, where possible, come up with different ways of
      encoding them in morphosyntax than I am familiar with in natlangs.
      Finally, a practical application for my formal semantics course! (Heh.
      "Practical." Well, if it makes me happy, that's practical, right?)

      Although there's a good deal of reference to math & logic in the
      design process, this is *not* intended to be a loglang, nor a
      minimalist-lang. It should feel complicated and lived in, with lots of
      dials that could be fiddled, so that it could plausibly be conceived
      of as a member of a larger family of languages that are all kinda
      weird but do in fact have the dials set differently from this one.

      I had the idea to use Armenian with some sound changes as an initial
      vocabulary source; since I don't actually know Armenian, it should
      quickly become unrecognizable. Unfortunately, I then realized I had
      already packed up my notebooks of Armenian elicitation data so I can
      move apartments in a few days! So, actual language samples are gonna
      have to wait a bit. For now, just more glosses.

      The basic structure of individually specifying the role of every
      participant in a clause is similar to what I had done with Gogido
      already; Gogido, however, actually does have verbs, and merely allows
      them to be elided sometimes, where WSL doesn't have verbs to elide in
      the first place. Still, I can steal some expressive tricks from my
      pre-existing thoughts on Gogido.

      So, here's what I've got so far:

      The basic parts of speech are Projectors (like complementizers, but
      present in all clauses, not just complements), Modals, Nouns (content
      words, but shorter to type), Roles (postpositions, but shorter to
      type), Conjunctions, and Connectives. 6, so far. Or 5, if you group
      conjunctions and connectives together; the difference is that
      connectives can only act on clauses with truth values, while
      connectives operate on constituents that denote entities.

      WSL nouns generally represent one-place predicates (covering English
      nouns, adjectives, adverbs, and intransitive verbs), while roles
      represent two-place predicates. Some English nouns and adjectives do
      not actually have the semantics of 1-place predicates; kinship terms,
      for example, are all defined by a relation between *two* entities.
      There are a couple of ways I could handle that. Maybe the word for
      "father" is actually a role, rather than a noun. Or maybe the
      equivalent for "father" is a one-place neo-Davidsonian-curried
      predicate denoting the relation of fatherhood, which has to have the
      participants specified separately by other roles. Since this isn't a
      loglang, I don't have to be consistent! I can do either or both on a
      concept-by-concept basis.

      Roles, unlike regular postpositions, are going to be an open class
      like nouns. This is mainly because I don't expect to be able to
      enumerate all of the possible semantic roles that might ever be
      relevant to anything; so, there'll be a core set of basic,
      unanalyzable role words, and then some morphological process to derive
      new ones from nouns.

      Modals are pretty much what they sound like- they tell you the mood of
      the clause. And I'm thinking I might throw some evidentiality into the
      mix there, too.

      Clauses come in multiple types, depending on whether they project /
      denote a truth value (top-level clauses making sentences must project
      truth values), the proposition represented by the clause (analogous to
      a typical complement clause), or one its arguments.

      Noun phrases consist of single nouns or nominal clauses (any clause
      that isn't a truth-valued sentence), or two simpler noun phrases
      joined by a conjunction. Two noun phrases in hiatus are analyzed as
      being joined by the null conjunction, which performs an intersection /
      logical 'and' operation on their denotations. Thus, WSL "red apple"
      (subbing in WSL words for English glosses later) and "apple red" both
      mean English "red apple"- something that is a red thing and is also an
      apple. Thus, we can translate complicated English noun phrases with
      lots of adjectives and complicated verb phrases with lots of adverbs
      very simply.

      Clauses generally consist of a projector, an optional specifier, some
      number of argument phrases, and a modal. The projector and the modal
      have to come at the beginning and end, respectively, but in between
      things can be in any order.

      Argument phrases are noun phrases, followed by an optional topic
      clitic (to be explained later) and at least one role. Specifiers are
      noun phrases followed by a specifier clitic.

      Every clause has an implicit "there exist some event/relation/entity
      such that I can plug it into all of the roles in this clause and have
      the proposition bet true"; in the initial sketch, I had an EVT role
      that defined what that event was. However, I decided that was
      inelegant because it would be the only one-place role; so, it gets
      demoted to the specifier clitic, and the purpose of the optional
      specifier is to tell you what the relation/entity is that all of the
      roles relate to. This could be done with word order instead of
      introducing morphological marking, but I have other things in mind
      that will require being able to fiddle with the phrase order later, so
      for now it's left totally free. Specifiers can also contain the topic
      clitic.

      So far, there are four projectors: positive sentential, negative
      sentential, positive entity, and negative entity. These control both
      whether the clause denotes a truth-valued sentence or not, and also
      whether the clause has positive or negative polarity. This represents
      another change from the original sketch, where I had NEG as a
      connective; I decided NEG didn't really fit, as the only unary
      connective, so its semantics get merged into the projectors.

      Existential and copular sentences can be translated as WSL sentences
      with a specifier, which may or may not have any arguments.

      SEN.POS dog-SPEC DEC
      "There exists something which is a dog (declarative)" / "There's a dog".

      SEN.POS dog furry the-SPEC DEC
      "There exists something which is the dog and which is a furry" / "The
      dog is furry."
      I'm kind of glossing over how translation of articles works here, but
      for suffice it to say that articles are just semantically-odd WSL
      nouns, just like English nouns, adjective, verbs, and adverbs.

      SEN.NEG me woman-SPEC DEC
      "There does not exist a thing that is me and that is a woman." / "I am
      not a woman."

      SEN.POS mouse-SPEC sink LOC INT
      "Does there exist something which is a mouse and whose location is a
      sink?" / "Is there a mouse in the sink?"

      Regular sentences with meaningful verbs are translated exactly the
      same way; as far as WSL is concerned, there is no difference in
      structure, just a difference in the semantics of the noun you choose
      to put in the specifier slot.

      SEN.POS me EXP like-SPEC read THM DEC
      "I like to read." / "I like reading."

      Multiple argument phrases in a clause can use the same role, in which
      case their denotations are unioned.

      SEN.POS me EXP like-SPEC read THM write THM DEC
      "I like to read and write." / "I like reading and writing."

      The same effect could be accomplished with an appropriate conjunction
      in a single argument phrase, but this is a useful mechanism for
      breaking up long, complex argument phrases that may have embedded
      clauses to make them easier to handle.

      I'm waffling on whether or not to allow the topic clitic in sentence
      clauses. If I do, it would probably indicate topic (just like the name
      implies!) or some kind of focus.

      Complement clauses and relative clauses are translated with the entity
      projectors. If a topic is marked in the clause, then it behaves like
      an internally-headed relative clause. Otherwise, it behaves like a
      complement clause. In some cases, English sentences can be translated
      either way:

      SEN.POS me EXP ENT.POS ride-SPEC bikes PAT DEC THM like-SPEC DEC
      "I like bike riding." / "I like that bikes are ridden." (This turns
      passive in English because the WSL didn't specify an agent.)

      SEN.POS me EXP ENT.POS ride-TOP-SPEC bikes PAT DEC THM like-SPEC DEC
      "I like bike riding." / "I like the riding of bikes."

      Compare:

      SEN.POS me EXP ENT.POS ride-SPEC bikes-TOP PAT DEC THM like-SPEC DEC
      "I like ridden bikes."

      Or perhaps the more sensible:

      SEN.POS ENT.POS bag-TOP PAT open-SPEC DEC PAT me LOC DEC
      "I have an open bag." / "A bag which is open is at me."

      Note that this also demonstrates the use of a sentence with no specifier.

      Reflexives can be indicated by using multiple roles in a single argument phrase.

      SEN.POS hit-SPEC me AG PAT DEC
      "I hit myself."

      And reciprocals can be indicated by using the same set of roles in
      differing orders on different arguments.

      SEN.POS hit-SPEC me AG PAT him PAT AG DEC
      "He and I hit each other."

      Note that this requires two explicit arguments to be present to form a
      reciprocal; there's no way to form a reciprocal from a single argument
      that has a plural denotation; e.g., "We hit each other." Might think
      some more about how to do that nicely, or just decide that the
      language doesn't need it.

      I think that's all for now. So far, you may have noticed me taking
      advantage of the free phrase order to avoid inelegant pileups like
      "PAT DEC PAT DEC" at the end of a sentence. Order will become
      important when I get around to introducing quantifiers, though- I
      think I can arrange for totally unambiguous quantifier scope. I have
      also been thinking about figuring out some clefting or dislocation
      mechanism that could help reduce clause embedding, also reducing the
      occurrence of particle pileups.

      -l.
    • J S Jones
      ... This is where I get lost. Is _read_ a noun?
      Message 2 of 9 , Jul 28 5:42 AM
        On Sun, 27 Jul 2014 18:44:28 -0600, Logan Kearsley <chronosurfer@...> wrote:
        >
        >The basic parts of speech are Projectors (like complementizers, but
        >present in all clauses, not just complements), Modals, Nouns (content
        >words, but shorter to type), Roles (postpositions, but shorter to
        >type), Conjunctions, and Connectives. 6, so far. Or 5, if you group
        >conjunctions and connectives together; the difference is that
        >connectives can only act on clauses with truth values, while
        >connectives operate on constituents that denote entities.
        >
        >WSL nouns generally represent one-place predicates (covering English
        >nouns, adjectives, adverbs, and intransitive verbs), while roles
        >represent two-place predicates. Some English nouns and adjectives do
        >not actually have the semantics of 1-place predicates; kinship terms,
        >for example, are all defined by a relation between *two* entities.
        >There are a couple of ways I could handle that. Maybe the word for
        >"father" is actually a role, rather than a noun. Or maybe the
        >equivalent for "father" is a one-place neo-Davidsonian-curried
        >predicate denoting the relation of fatherhood, which has to have the
        >participants specified separately by other roles. Since this isn't a
        >loglang, I don't have to be consistent! I can do either or both on a
        >concept-by-concept basis.
        >
        >Roles, unlike regular postpositions, are going to be an open class
        >like nouns. This is mainly because I don't expect to be able to
        >enumerate all of the possible semantic roles that might ever be
        >relevant to anything; so, there'll be a core set of basic,
        >unanalyzable role words, and then some morphological process to derive
        >new ones from nouns.
        >
        >Modals are pretty much what they sound like- they tell you the mood of
        >the clause. And I'm thinking I might throw some evidentiality into the
        >mix there, too.
        >
        >Clauses come in multiple types, depending on whether they project /
        >denote a truth value (top-level clauses making sentences must project
        >truth values), the proposition represented by the clause (analogous to
        >a typical complement clause), or one its arguments.
        >
        >Noun phrases consist of single nouns or nominal clauses (any clause
        >that isn't a truth-valued sentence), or two simpler noun phrases
        >joined by a conjunction. Two noun phrases in hiatus are analyzed as
        >being joined by the null conjunction, which performs an intersection /
        >logical 'and' operation on their denotations. Thus, WSL "red apple"
        >(subbing in WSL words for English glosses later) and "apple red" both
        >mean English "red apple"- something that is a red thing and is also an
        >apple. Thus, we can translate complicated English noun phrases with
        >lots of adjectives and complicated verb phrases with lots of adverbs
        >very simply.
        >
        >Clauses generally consist of a projector, an optional specifier, some
        >number of argument phrases, and a modal. The projector and the modal
        >have to come at the beginning and end, respectively, but in between
        >things can be in any order.
        >
        >Argument phrases are noun phrases, followed by an optional topic
        >clitic (to be explained later) and at least one role. Specifiers are
        >noun phrases followed by a specifier clitic.
        >
        >Every clause has an implicit "there exist some event/relation/entity
        >such that I can plug it into all of the roles in this clause and have
        >the proposition bet true"; in the initial sketch, I had an EVT role
        >that defined what that event was. However, I decided that was
        >inelegant because it would be the only one-place role; so, it gets
        >demoted to the specifier clitic, and the purpose of the optional
        >specifier is to tell you what the relation/entity is that all of the
        >roles relate to. This could be done with word order instead of
        >introducing morphological marking, but I have other things in mind
        >that will require being able to fiddle with the phrase order later, so
        >for now it's left totally free. Specifiers can also contain the topic
        >clitic.
        >
        >So far, there are four projectors: positive sentential, negative
        >sentential, positive entity, and negative entity. These control both
        >whether the clause denotes a truth-valued sentence or not, and also
        >whether the clause has positive or negative polarity. This represents
        >another change from the original sketch, where I had NEG as a
        >connective; I decided NEG didn't really fit, as the only unary
        >connective, so its semantics get merged into the projectors.
        >
        >Existential and copular sentences can be translated as WSL sentences
        >with a specifier, which may or may not have any arguments.
        >
        >SEN.POS dog-SPEC DEC
        >"There exists something which is a dog (declarative)" / "There's a dog".
        >
        >SEN.POS dog furry the-SPEC DEC
        >"There exists something which is the dog and which is a furry" / "The
        >dog is furry."
        >I'm kind of glossing over how translation of articles works here, but
        >for suffice it to say that articles are just semantically-odd WSL
        >nouns, just like English nouns, adjective, verbs, and adverbs.
        >
        >SEN.NEG me woman-SPEC DEC
        >"There does not exist a thing that is me and that is a woman." / "I am
        >not a woman."
        >
        >SEN.POS mouse-SPEC sink LOC INT
        >"Does there exist something which is a mouse and whose location is a
        >sink?" / "Is there a mouse in the sink?"
        >
        >Regular sentences with meaningful verbs are translated exactly the
        >same way; as far as WSL is concerned, there is no difference in
        >structure, just a difference in the semantics of the noun you choose
        >to put in the specifier slot.
        >
        >SEN.POS me EXP like-SPEC read THM DEC
        >"I like to read." / "I like reading."

        This is where I get lost. Is _read_ a noun?

        >Multiple argument phrases in a clause can use the same role, in which
        >case their denotations are unioned.
        >
        >SEN.POS me EXP like-SPEC read THM write THM DEC
        >"I like to read and write." / "I like reading and writing."
        >
        >The same effect could be accomplished with an appropriate conjunction
        >in a single argument phrase, but this is a useful mechanism for
        >breaking up long, complex argument phrases that may have embedded
        >clauses to make them easier to handle.
        >
        >I'm waffling on whether or not to allow the topic clitic in sentence
        >clauses. If I do, it would probably indicate topic (just like the name
        >implies!) or some kind of focus.
        >
        >Complement clauses and relative clauses are translated with the entity
        >projectors. If a topic is marked in the clause, then it behaves like
        >an internally-headed relative clause. Otherwise, it behaves like a
        >complement clause. In some cases, English sentences can be translated
        >either way:
        >
        >SEN.POS me EXP ENT.POS ride-SPEC bikes PAT DEC THM like-SPEC DEC
        >"I like bike riding." / "I like that bikes are ridden." (This turns
        >passive in English because the WSL didn't specify an agent.)
        >
        >SEN.POS me EXP ENT.POS ride-TOP-SPEC bikes PAT DEC THM like-SPEC DEC
        >"I like bike riding." / "I like the riding of bikes."
        >
        >Compare:
        >
        >SEN.POS me EXP ENT.POS ride-SPEC bikes-TOP PAT DEC THM like-SPEC DEC
        >"I like ridden bikes."
        >
        >Or perhaps the more sensible:
        >
        >SEN.POS ENT.POS bag-TOP PAT open-SPEC DEC PAT me LOC DEC
        >"I have an open bag." / "A bag which is open is at me."
        >
        >Note that this also demonstrates the use of a sentence with no specifier.
        >
        >Reflexives can be indicated by using multiple roles in a single argument phrase.
        >
        >SEN.POS hit-SPEC me AG PAT DEC
        >"I hit myself."
        >
        >And reciprocals can be indicated by using the same set of roles in
        >differing orders on different arguments.
        >
        >SEN.POS hit-SPEC me AG PAT him PAT AG DEC
        >"He and I hit each other."
        >
        >Note that this requires two explicit arguments to be present to form a
        >reciprocal; there's no way to form a reciprocal from a single argument
        >that has a plural denotation; e.g., "We hit each other." Might think
        >some more about how to do that nicely, or just decide that the
        >language doesn't need it.
        >
        >
        >-l.
      • Logan Kearsley
        ... [...] SEN.POS me EXP like-SPEC read THM DEC ... Yes. me , like , and read are all nouns in this sentence. An unfortunate limitation of glossing in
        Message 3 of 9 , Jul 28 6:54 AM
          On 28 July 2014 06:42, J S Jones <qiihoskeh@...> wrote:
          > On Sun, 27 Jul 2014 18:44:28 -0600, Logan Kearsley <chronosurfer@...> wrote:
          [...]>>SEN.POS me EXP like-SPEC read THM DEC
          >>"I like to read." / "I like reading."
          >
          > This is where I get lost. Is _read_ a noun?

          Yes. "me", "like", and "read" are all nouns in this sentence.
          An unfortunate limitation of glossing in English is that it results in
          confusion with English categories. I'm not sure how much it will help
          other people, since English glossing is still required, but this is
          why I like to use other language as vocab sources for experiments like
          this. Which I'll get around to after I unpack my notebooks.

          -l.
        • Garth Wallace
          ... . . . ... What about sentences that do not make an assertion? Like that one?
          Message 4 of 9 , Jul 28 2:10 PM
            On Sun, Jul 27, 2014 at 5:44 PM, Logan Kearsley <chronosurfer@...> wrote:
            >
            > Clauses come in multiple types, depending on whether they project /
            > denote a truth value (top-level clauses making sentences must project
            > truth values), the proposition represented by the clause (analogous to
            > a typical complement clause), or one its arguments.
            .
            .
            .
            > Every clause has an implicit "there exist some event/relation/entity
            > such that I can plug it into all of the roles in this clause and have
            > the proposition bet true"; in the initial sketch, I had an EVT role
            > that defined what that event was. However, I decided that was
            > inelegant because it would be the only one-place role; so, it gets
            > demoted to the specifier clitic, and the purpose of the optional
            > specifier is to tell you what the relation/entity is that all of the
            > roles relate to. This could be done with word order instead of
            > introducing morphological marking, but I have other things in mind
            > that will require being able to fiddle with the phrase order later, so
            > for now it's left totally free. Specifiers can also contain the topic
            > clitic.
            >
            > So far, there are four projectors: positive sentential, negative
            > sentential, positive entity, and negative entity. These control both
            > whether the clause denotes a truth-valued sentence or not, and also
            > whether the clause has positive or negative polarity. This represents
            > another change from the original sketch, where I had NEG as a
            > connective; I decided NEG didn't really fit, as the only unary
            > connective, so its semantics get merged into the projectors.

            What about sentences that do not make an assertion? Like that one?
          • Logan Kearsley
            ... At the moment, that s handled by the modal particle. I had one examples for that, using interrogative mood instead of declarative: SEN.POS mouse-SPEC sink
            Message 5 of 9 , Jul 28 2:48 PM
              On 28 July 2014 15:10, Garth Wallace <gwalla@...> wrote:
              > On Sun, Jul 27, 2014 at 5:44 PM, Logan Kearsley <chronosurfer@...> wrote:
              >>
              >> Clauses come in multiple types, depending on whether they project /
              >> denote a truth value (top-level clauses making sentences must project
              >> truth values), the proposition represented by the clause (analogous to
              >> a typical complement clause), or one its arguments.
              > .
              > .
              > .
              >> Every clause has an implicit "there exist some event/relation/entity
              >> such that I can plug it into all of the roles in this clause and have
              >> the proposition bet true"; in the initial sketch, I had an EVT role
              >> that defined what that event was. However, I decided that was
              >> inelegant because it would be the only one-place role; so, it gets
              >> demoted to the specifier clitic, and the purpose of the optional
              >> specifier is to tell you what the relation/entity is that all of the
              >> roles relate to. This could be done with word order instead of
              >> introducing morphological marking, but I have other things in mind
              >> that will require being able to fiddle with the phrase order later, so
              >> for now it's left totally free. Specifiers can also contain the topic
              >> clitic.
              >>
              >> So far, there are four projectors: positive sentential, negative
              >> sentential, positive entity, and negative entity. These control both
              >> whether the clause denotes a truth-valued sentence or not, and also
              >> whether the clause has positive or negative polarity. This represents
              >> another change from the original sketch, where I had NEG as a
              >> connective; I decided NEG didn't really fit, as the only unary
              >> connective, so its semantics get merged into the projectors.
              >
              > What about sentences that do not make an assertion? Like that one?

              At the moment, that's handled by the modal particle. I had one
              examples for that, using interrogative mood instead of declarative:

              SEN.POS mouse-SPEC sink LOC INT
              "Is there a mouse in the sink?"

              That might end up changing, though. Adding another two projectors for
              positive/negative questions might be a better solution, since
              questioning is not really orthogonal with other moods; i.e., you can
              make something subjunctive and a question at the same time, for
              example.

              My model for designing the semantics of modals right now is based on
              treating them as quantifiers over possible worlds- what percentage of
              possible worlds is this true in / what's the probability that it's
              true, how close are the worlds described by this proposition to what I
              think is the real world, etc. That should represent all of the modal
              concepts we're familiar with from English, etc. (because that
              formulation comes from formal semantic models developed for English,
              after all), but not quite in exactly the same way. Concepts relating
              to desirability and personal opinions on situations (like "should")
              might be mixed in as well, or I might end up lexicalizing those as
              nouns in order to keep it Weird. Either way, interrogativity doesn't
              really fit with that model.

              There's also imperatives to think about. I am strongly considering
              lexicalizing that, though, rather than putting imperatives into the
              grammar. Or just leaving it up to pragmatics- after all, what
              difference is there *really* between "Clean your room!" (not an
              assertion) and "You should be cleaning your room!" (definitely an
              assertion).

              So, yeah, I'm thinking I should ditch "interrogative mood" and end up
              with 6-8 projectors instead of 4.

              -l.
            • Alex Fink
              I mean to read your later description of WSL (with words!) at some point, but for now just a tangential question: ... Are you aware of any formal semantic
              Message 6 of 9 , Aug 27, 2014
                I mean to read your later description of WSL (with words!) at some point, but for now just a tangential question:

                On Mon, 28 Jul 2014 15:48:30 -0600, Logan Kearsley <chronosurfer@...> wrote:

                >My model for designing the semantics of modals right now is based on
                >treating them as quantifiers over possible worlds- what percentage of
                >possible worlds is this true in / what's the probability that it's
                >true, how close are the worlds described by this proposition to what I
                >think is the real world, etc. That should represent all of the modal
                >concepts we're familiar with from English, etc. (because that
                >formulation comes from formal semantic models developed for English,
                >after all), but not quite in exactly the same way. Concepts relating
                >to desirability and personal opinions on situations (like "should")
                >might be mixed in as well, or I might end up lexicalizing those as
                >nouns in order to keep it Weird. Either way, interrogativity doesn't
                >really fit with that model.

                Are you aware of any formal semantic models (say, developed in a non-English tradition) that have a different suite of modal concepts? I'd be interested in such.

                In UNLWS, where quantifiers over possible worlds are a big deal, we have decided to demote the equivalents of precise percentage quantifiers to the number system and use them by dispreference to the quantifiers which express typicality; this is against what seems the general loglang preference for a "most == at least three quarters of" or what have you. We do though have quantifiers for unspecified small and large proportions, and one for "greater proportion than for the [implicit] super-category of the restrictum", attempting therein to remedy the way natlangs are usually bad at stating statistical claims. Other modals in UNLWS are underdeveloped.

                Alex
              • Logan Kearsley
                ... I am not. Different sources come at it with slightly different approaches, but everything I ve read about modality eventually comes around to the
                Message 7 of 9 , Aug 27, 2014
                  On 27 August 2014 17:39, Alex Fink <000024@...> wrote:
                  > I mean to read your later description of WSL (with words!) at some point, but for now just a tangential question:
                  >
                  > On Mon, 28 Jul 2014 15:48:30 -0600, Logan Kearsley <chronosurfer@...> wrote:
                  >
                  >>My model for designing the semantics of modals right now is based on
                  >>treating them as quantifiers over possible worlds- what percentage of
                  >>possible worlds is this true in / what's the probability that it's
                  >>true, how close are the worlds described by this proposition to what I
                  >>think is the real world, etc. That should represent all of the modal
                  >>concepts we're familiar with from English, etc. (because that
                  >>formulation comes from formal semantic models developed for English,
                  >>after all), but not quite in exactly the same way. Concepts relating
                  >>to desirability and personal opinions on situations (like "should")
                  >>might be mixed in as well, or I might end up lexicalizing those as
                  >>nouns in order to keep it Weird. Either way, interrogativity doesn't
                  >>really fit with that model.
                  >
                  > Are you aware of any formal semantic models (say, developed in a non-English tradition) that have a different suite of modal concepts? I'd be interested in such.

                  I am not. Different sources come at it with slightly different
                  approaches, but everything I've read about modality eventually comes
                  around to the possible-worlds interpretation.

                  The interesting bit is the expression of modal bases- i.e., what is
                  the domain over which quantification occurs, which defines the type of
                  modality (alethic, deontic, epistemic, etc.). Although every analysis
                  of modality seems to come around to possible world semantics, how the
                  modal base is expressed varies greatly both between and within
                  languages- it can be left to discourse context, or lexicalized in a
                  bunch of different places (modal verbs, matrix verbs, adverbs, special
                  particles like evidentials, etc.). I could get a compromise between
                  including expressions of feeling or opinion as part of the grammatical
                  mood system vs. lexicalizing them by introducing a role marker (or
                  some other construct) that says "this group of predicates / WSL Noun
                  phrase constrains the modal base of this proposition". Which
                  essentially allows for inventing arbitrary new kinds of modality
                  on-the-fly. No restriction gets you default alethic modality,
                  restriction with a predicate indicating the set of all things you
                  believe creates epistemic or doxastic modality, restriction with a
                  predicate indicating the set of all things you are OK with gets you
                  deontic modality, etc.

                  > In UNLWS, where quantifiers over possible worlds are a big deal, we have decided to demote the equivalents of precise percentage quantifiers to the number system and use them by dispreference to the quantifiers which express typicality; this is against what seems the general loglang preference for a "most == at least three quarters of" or what have you. We do though have quantifiers for unspecified small and large proportions, and one for "greater proportion than for the [implicit] super-category of the restrictum", attempting therein to remedy the way natlangs are usually bad at stating statistical claims. Other modals in UNLWS are underdeveloped.

                  I am aware of the idea of improving statistical language, but it's
                  been a while since I read or thought about that subject specifically,
                  so I'm having trouble wrapping my head around that last example. Could
                  you perhaps explain differently what "greater proportion than for the
                  [implicit] super-category of the restrictum" means and how that helps
                  with making statistical claims?

                  -l.
                • Alex Fink
                  ... Ah, interesting. We d only used the alethic version so far, although the phonological forms of be good and be expected are of an easily combinable
                  Message 8 of 9 , Aug 30, 2014
                    On Wed, 27 Aug 2014 18:46:20 -0600, Logan Kearsley <chronosurfer@...> wrote:

                    >On 27 August 2014 17:39, Alex Fink <000024@...> wrote:
                    >> I mean to read your later description of WSL (with words!) at some point, but for now just a tangential question:
                    >>
                    >> Are you aware of any formal semantic models (say, developed in a non-English tradition) that have a different suite of modal concepts? I'd be interested in such.
                    >
                    >I am not. Different sources come at it with slightly different
                    >approaches, but everything I've read about modality eventually comes
                    >around to the possible-worlds interpretation.
                    >
                    >The interesting bit is the expression of modal bases- i.e., what is
                    >the domain over which quantification occurs, which defines the type of
                    >modality (alethic, deontic, epistemic, etc.). Although every analysis
                    >of modality seems to come around to possible world semantics, how the
                    >modal base is expressed varies greatly both between and within
                    >languages- it can be left to discourse context, or lexicalized in a
                    >bunch of different places (modal verbs, matrix verbs, adverbs, special
                    >particles like evidentials, etc.). I could get a compromise between
                    >including expressions of feeling or opinion as part of the grammatical
                    >mood system vs. lexicalizing them by introducing a role marker (or
                    >some other construct) that says "this group of predicates / WSL Noun
                    >phrase constrains the modal base of this proposition". Which
                    >essentially allows for inventing arbitrary new kinds of modality
                    >on-the-fly. No restriction gets you default alethic modality,
                    >restriction with a predicate indicating the set of all things you
                    >believe creates epistemic or doxastic modality, restriction with a
                    >predicate indicating the set of all things you are OK with gets you
                    >deontic modality, etc.

                    Ah, interesting. We'd only used the alethic version so far, although the phonological forms of "be good" and "be expected" are of an easily combinable sort, so that it's compelling to find a way to work them in and get, respectively, deontic modality and, um, further alethic or maybe epistemic modality?

                    On the other hand... mentioning this to Sai just now, he doesn't appear to really be happy with deontic modality as a category at all. He wants to cast deontic "X must Y" as '[explicitly mentioned deemer] only finds the situation good in which X does Y', or maybe 'X will only achieve [explicitly mentioned goal] if they do Y', and regarding "X may Y" his immediate reaction was "who are you to give permission?!" and his suggestion '[explicitly mentioned deemer] will avoid visiting retribution on X if they do Y' (which to me is redolent of his very legalistic mindset), though he concedes that perhaps these should have abbreviations or allow "common default" values of the modal restriction vel sim. And the 'if' in these constructions hides an alethic modal construction in your sense, so perhaps via the abbreviation we'll end up with more than just alethic modality after all.

                    >> In UNLWS, [...] we do though have quantifiers for unspecified small and large proportions, and one for "greater proportion than for the [implicit] super-category of the restrictum", attempting therein to remedy the way natlangs are usually bad at stating statistical claims. Other modals in UNLWS are underdeveloped.
                    >
                    >I am aware of the idea of improving statistical language, but it's
                    >been a while since I read or thought about that subject specifically,
                    >so I'm having trouble wrapping my head around that last example. Could
                    >you perhaps explain differently what "greater proportion than for the
                    >[implicit] super-category of the restrictum" means and how that helps
                    >with making statistical claims?

                    Well, it's only the first step towards such an improvement, and we haven't thought very much about its systematic deployment or its congeners; perhaps you'll give us some other ideas. But it's meant to be used for reporting e.g. results like those graphed in http://itre.cis.upenn.edu/~myl/languagelog/archives/004965.html -- using this quantifier, glossed CORR (for "correlate"), one can say "women CORR are happy" meaning 'a greater proportion of women are happy than [implicitly] people are', which is marked by contrast to "women GEN are happy" ("generic"), read as 'women are happy' with the generic plural, or "women many are happy" read as expected, or others.

                    Alex
                  • Logan Kearsley
                    ... Yeah, just about every modality can be handled by periphrasis of that sort (after all, there must be some way to describe them for providing dictionary
                    Message 9 of 9 , Aug 31, 2014
                      On 30 August 2014 13:47, Alex Fink <000024@...> wrote:
                      > On Wed, 27 Aug 2014 18:46:20 -0600, Logan Kearsley <chronosurfer@...> wrote:
                      >
                      >>On 27 August 2014 17:39, Alex Fink <000024@...> wrote:
                      >>> I mean to read your later description of WSL (with words!) at some point, but for now just a tangential question:
                      >>>
                      >>> Are you aware of any formal semantic models (say, developed in a non-English tradition) that have a different suite of modal concepts? I'd be interested in such.
                      >>
                      >>I am not. Different sources come at it with slightly different
                      >>approaches, but everything I've read about modality eventually comes
                      >>around to the possible-worlds interpretation.
                      >>
                      >>The interesting bit is the expression of modal bases- i.e., what is
                      >>the domain over which quantification occurs, which defines the type of
                      >>modality (alethic, deontic, epistemic, etc.). Although every analysis
                      >>of modality seems to come around to possible world semantics, how the
                      >>modal base is expressed varies greatly both between and within
                      >>languages- it can be left to discourse context, or lexicalized in a
                      >>bunch of different places (modal verbs, matrix verbs, adverbs, special
                      >>particles like evidentials, etc.). I could get a compromise between
                      >>including expressions of feeling or opinion as part of the grammatical
                      >>mood system vs. lexicalizing them by introducing a role marker (or
                      >>some other construct) that says "this group of predicates / WSL Noun
                      >>phrase constrains the modal base of this proposition". Which
                      >>essentially allows for inventing arbitrary new kinds of modality
                      >>on-the-fly. No restriction gets you default alethic modality,
                      >>restriction with a predicate indicating the set of all things you
                      >>believe creates epistemic or doxastic modality, restriction with a
                      >>predicate indicating the set of all things you are OK with gets you
                      >>deontic modality, etc.
                      >
                      > Ah, interesting. We'd only used the alethic version so far, although the phonological forms of "be good" and "be expected" are of an easily combinable sort, so that it's compelling to find a way to work them in and get, respectively, deontic modality and, um, further alethic or maybe epistemic modality?
                      >
                      > On the other hand... mentioning this to Sai just now, he doesn't appear to really be happy with deontic modality as a category at all. He wants to cast deontic "X must Y" as '[explicitly mentioned deemer] only finds the situation good in which X does Y', or maybe 'X will only achieve [explicitly mentioned goal] if they do Y', and regarding "X may Y" his immediate reaction was "who are you to give permission?!" and his suggestion '[explicitly mentioned deemer] will avoid visiting retribution on X if they do Y' (which to me is redolent of his very legalistic mindset), though he concedes that perhaps these should have abbreviations or allow "common default" values of the modal restriction vel sim. And the 'if' in these constructions hides an alethic modal construction in your sense, so perhaps via the abbreviation we'll end up with more than just alethic modality after all.

                      Yeah, just about every modality can be handled by periphrasis of that
                      sort (after all, there must be some way to describe them for providing
                      dictionary definitions!).
                      To assist in thinking about language design, I've been using a
                      somewhat more formulaic periphrasis, such as I could put in a gloss.
                      E.g., "Given X, ..." or "X being true implies...", where "X" is a set
                      of propositions.

                      Epistemic/Doxastic modality is periphrased as "Given that I know..." /
                      "Given my beliefs..." or "The truth of my beliefs implies..."
                      Deontic modality is "Given that someone approves of that outcome...."
                      or "That someone will approve of the world is true implies...."

                      Of course, more complicated and situationally specific periphrases are
                      possible as you quote Sai giving for deontic universal (must) and
                      existential (may) cases, but those are the sorts of
                      generally-applicable formulas that I might expect to become naturally
                      grammaticalized.

                      One issue I've been mentally wrestling with in my current conception
                      of how WSL modality should work is that things like doxastic modality
                      require predicates / WSL Nouns for collections of proposition /
                      worlds. Reifying those things into the language is fine, but I wonder
                      how frequently they would actually be used in any other context; how
                      often do I really need a word that means "all of the possible worlds
                      in which everything I believe is true"? Those formulaic periphrases
                      are already totally possible in WSL as defined so far, so I could give
                      up on grammaticalizing modal basis entirely, except that

                      1. I really want to keep around the idea of clause-final modal phrases
                      for structural / parsing reasons
                      2. Unless I want to construct a language which always requires
                      explicitly stating your assumptions and presuppositions behind any
                      statement (which might be a bad thing for a loglang, but is not
                      aesthetically what I'm going for here), you still need those weird
                      "collection of worlds" words to plug in to the formulaic phrases, and
                      I kinda feel like those are just begging for grammaticalization.

                      So, perhaps I have a bunch of predefined modal base words that are
                      just indivisible particles or category "Mood" (kinda like proper
                      nouns), but then also allow for complex modal phrases when you
                      actually do want to make assumptions explicit (like common noun
                      phrases).

                      Another thing I've been thinking about is whether other quantifier
                      types besides basic must/may and continuum of probability between them
                      can actually make sense in modal constructions, because I feel it
                      would be elegant if one common quantifier construct can be re-used
                      between WSL Noun phrases and Mood phrases. E.g., "at least 3 but no
                      more than 50%" make sense as a complex quantifier when referring to,
                      say, chocolate bars, but "in at least 3 possible worlds"? To make
                      things work, I'm thinking that discrete quantification like that is
                      interpreted in the modal context as an indicator of a number of
                      possible explanations. So, "in at least 3 but no more than 50% of
                      possible worlds" would *actually* mean something like "I can think of
                      3 ways that this could happen / 3 situations in which this must be
                      true, but there could be more up to 50% total probability".

                      Of course, I still have to work out exactly how I want to handle
                      complex quantifiers in WSL at all....

                      >>> In UNLWS, [...] we do though have quantifiers for unspecified small and large proportions, and one for "greater proportion than for the [implicit] super-category of the restrictum", attempting therein to remedy the way natlangs are usually bad at stating statistical claims. Other modals in UNLWS are underdeveloped.
                      >>
                      >>I am aware of the idea of improving statistical language, but it's
                      >>been a while since I read or thought about that subject specifically,
                      >>so I'm having trouble wrapping my head around that last example. Could
                      >>you perhaps explain differently what "greater proportion than for the
                      >>[implicit] super-category of the restrictum" means and how that helps
                      >>with making statistical claims?
                      >
                      > Well, it's only the first step towards such an improvement, and we haven't thought very much about its systematic deployment or its congeners; perhaps you'll give us some other ideas. But it's meant to be used for reporting e.g. results like those graphed in http://itre.cis.upenn.edu/~myl/languagelog/archives/004965.html -- using this quantifier, glossed CORR (for "correlate"), one can say "women CORR are happy" meaning 'a greater proportion of women are happy than [implicitly] people are', which is marked by contrast to "women GEN are happy" ("generic"), read as 'women are happy' with the generic plural, or "women many are happy" read as expected, or others.

                      Ah, I get it now. This is something I feel like I have discussed
                      before. Perhaps a more 'robust' gloss could be something like "a
                      statistically-significantly larger fraction of women are happy than
                      you would naively assume"? With the option of adding periphrastic
                      explanations of what the type and level of significance is, as well as
                      what the relevant prior assumptions are. That generalizes that
                      possible implicit contextual inputs, rather than relying specifically
                      on a not-necessarily-unique supercategory.

                      -l.
                    Your message has been successfully submitted and would be delivered to recipients shortly.