Documentation

Linglib.Theories.Semantics.Questions.Answerhood.MentionSome

Questions/MentionSome.lean #

@cite{belnap-1982} @cite{groenendijk-stokhof-1984} @cite{partee-rooth-1983}

Mention-Some Interpretation from @cite{groenendijk-stokhof-1984}, Chapter VI, Section 5.

The Phenomenon #

"Where can I buy an Italian newspaper?" can be answered with a single location, even though there may be many places selling Italian newspapers. This is the mention-some reading, in contrast to the mention-all reading that would require listing all such locations.

Section 5.2: Partial Answerhood #

G&S first attempt to capture mention-some via partial answerhood (P-ANS):

P-ANS(p, q) ≡ ∃i[p(i) ∧ q(a)(i)] ∧ ¬∀a'∃i[p(i) ≡ q(a')(i)]

However, this fails because negative partial answers satisfy P-ANS but are NOT acceptable mention-some answers:

Section 5.3: The I-MS Rule #

The proper treatment requires a special interrogative formation rule:

I-MS: λQ[∃x[β'(x) ∧ Q(λaλi[β'(x) = (λxβ')(a)(i)])]]

This creates a lifted interrogative that takes a property of questions and returns a proposition. The mention-some reading is built into the semantics.

Embedded Mention-Some #

Under "know" (extensional): John knows who has a pen =

∃x[has-pen(x) ∧ know*(j, has-pen(x))]

Under "wonder" (intensional): John wonders who has a pen =

want(j, ∃x[has-pen(x) ∧ know*(j, has-pen(x))])

Licensing Factors (Section 5.4) #

Mention-some is licensed by:

  1. Human concerns (goals the questioner can achieve with partial information)
  2. Wide-scope existentials
  3. Some verbs block mention-some: "depends", "matter", "determine"

Partial Answerhood #

@cite{groenendijk-stokhof-1984}, Section 5.2: A proposition p gives a partial answer to question q if:

  1. p is compatible with some cell of q (overlaps with the answer)
  2. p is NOT a complete answer (doesn't determine a unique cell)

This captures the intuition that "Not in the drawer" partially answers "Where is the pen?" but fails to capture why it's not a mention-some answer.

def Semantics.Questions.MentionSome.partialAnswer {W : Type u_1} (p : WBool) (q : GSQuestion W) (worlds : List W) :

Partial answerhood: p overlaps with some cell but doesn't determine a unique cell.

@cite{groenendijk-stokhof-1984}, p. 335: P-ANS(p, q) ≡ ∃i[p(i) ∧ q(a)(i)] ∧ ¬∀a'∃i[p(i) ≡ q(a')(i)]

Equations
Instances For
    def Semantics.Questions.MentionSome.isPositivePartialAnswer {E : Type u_1} [BEq E] (answer : E) (satisfiers : List E) :

    Positive partial answer: mentions at least one satisfier.

    @cite{groenendijk-stokhof-1984}, Section 5.2: The problem with P-ANS is that negative answers like "Not in the drawer" satisfy P-ANS but are NOT acceptable mention-some answers.

    Only POSITIVE partial answers (that mention actual satisfiers) count as mention-some.

    Equations
    Instances For
      theorem Semantics.Questions.MentionSome.partialAnswer_includes_negative :
      ∃ (W : Type) (q : GSQuestion W) (worlds : List W) (p : WBool), partialAnswer p q worlds = true ¬(worlds.any p = true (worlds.all fun (w : W) => p w == true) = true)

      Why partial answerhood fails to capture mention-some: Negative answers satisfy P-ANS but shouldn't be mention-some answers.

      Example:

      • Q: "Where is a pen?"
      • "Not in the drawer" - satisfies P-ANS (eliminates some possibilities)
      • "In the study" - satisfies P-ANS AND is a proper mention-some answer

      The difference: only positive answers that mention actual locations work. This theorem witnesses a negative proposition that satisfies partialAnswer. [sorry: need concrete countermodel with worlds and question]

      The I-MS Interrogative Formation Rule #

      @cite{groenendijk-stokhof-1984}, Section 5.3: The proper treatment of mention-some requires a special interrogative formation rule that creates lifted interrogatives.

      Standard I-rule: λQ[∀x[P(x) → Q(x)]]

      I-MS rule: λQ[∃x[β'(x) ∧ Q(λaλi[β'(x) = (λxβ')(a)(i)])]]

      structure Semantics.Questions.MentionSome.MentionSomeInterrogative (W : Type u_1) (E : Type u_2) :
      Type (max u_1 u_2)

      A mention-some interrogative: the semantic structure created by I-MS.

      The whDomain lists possible values for the wh-phrase. The abstract β' is the predicate (e.g., "x has a pen" or "x sells newspapers").

      • whDomain : List E

        The wh-domain (possible values for the wh-phrase)

      • abstract : WEBool

        The wh-abstract β'(w, x): does x satisfy the property in world w?

      Instances For
        def Semantics.Questions.MentionSome.yesNoQuestionFor {W : Type u_1} {E : Type u_2} [DecidableEq E] (abstract : WEBool) (x : E) :

        Yes/no question for a specific value: "Does x satisfy β?"

        This is the inner question that the I-MS rule applies Q to.

        Equations
        Instances For

          The I-MS rule applied to a property of questions.

          I-MS: λQ[∃x[β'(x) ∧ Q(λaλi[β'(x) = (λxβ')(a)(i)])]]

          Given a property Q of questions (e.g., "John knows the answer to"), this returns true if there exists some x such that:

          1. x satisfies β' in the actual world (β'(x))
          2. Q holds of the yes/no question "does x satisfy β'?"
          Equations
          Instances For

            Embedded Mention-Some #

            @cite{groenendijk-stokhof-1984}, Section 5.3: Mention-some questions embedded under attitude verbs:

            Under "know" (extensional complement) #

            "John knows who has a pen" (mention-some) = ∃x[has-pen(x) ∧ know*(j, has-pen(x))]

            Paraphrase: "There is someone who has a pen and John knows that they have a pen"

            Under "wonder" (intensional complement) #

            "John wonders who has a pen" (mention-some) = want(j, ∃x[has-pen(x) ∧ know*(j, has-pen(x))])

            Paraphrase: "John wants there to be someone who has a pen and whom he knows has a pen" (i.e., John wants to find out about one pen-haver)

            def Semantics.Questions.MentionSome.knowMentionSome {W : Type u_1} {E : Type u_2} [DecidableEq E] (msi : MentionSomeInterrogative W E) (knows : WE(WBool)Bool) (agent : E) (w : W) :

            "Know" embedding of mention-some question (extensional reading).

            "John knows who has a pen" (mention-some) = ∃x[has-pen(x) ∧ know*(j, has-pen(x))]

            The agent knows the answer iff they know of SOME satisfier that it's a satisfier.

            Equations
            Instances For
              def Semantics.Questions.MentionSome.wonderMentionSome {W : Type u_1} {E : Type u_2} [DecidableEq E] (msi : MentionSomeInterrogative W E) (wants knows : WE(WBool)Bool) (agent : E) (w : W) :

              "Wonder" embedding of mention-some question (intensional reading).

              "John wonders who has a pen" (mention-some) = want(j, ∃x[has-pen(x) ∧ know*(j, has-pen(x))])

              The agent wonders Q iff they want to know of SOME satisfier that it's a satisfier.

              Equations
              • One or more equations did not get rendered due to their size.
              Instances For
                def Semantics.Questions.MentionSome.askMentionSome {W : Type u_1} {E : Type u_2} [DecidableEq E] (msi : MentionSomeInterrogative W E) (asks : WE(WBool)Bool) (agent : E) (w : W) :

                "Ask" embedding of mention-some question.

                "John asked who has a pen" (mention-some): John directed a question at some addressee, wanting to know of some pen-haver.

                Equations
                Instances For

                  Distinguishing Choice from Mention-Some #

                  @cite{groenendijk-stokhof-1984}, Section 5.1-5.3: Both choice and mention-some readings yield non-exhaustive answers, but they are semantically distinct:

                  Choice Reading: The disjunction/existential takes wide scope. "Whom does John or Mary love?" - Answer varies by who the lover is.

                  Mention-Some Reading: Goal-driven; any satisfier suffices. "Where can I buy coffee?" - Just need one place.

                  Key difference: Choice involves scope ambiguity; mention-some involves pragmatic goal-based interpretation.

                  What licenses mention-some readings.

                  Instances For
                    Equations
                    • One or more equations did not get rendered due to their size.
                    Instances For
                      structure Semantics.Questions.MentionSome.MentionSomeQuestion' (W : Type u_1) (E : Type u_2) :
                      Type (max u_1 u_2)

                      A mention-some question bundled with its reading type.

                      Instances For

                        Logical Relations Between Readings #

                        @cite{groenendijk-stokhof-1984}, Section 5.3 establishes:

                        1. Choice implies mention-some: If you know the choice-answer (for whichever disjunct is relevant), you know a mention-some answer.

                        2. Mention-all implies mention-some (if non-empty): If you know all satisfiers, you certainly know at least one.

                        These are important for understanding the logical landscape of readings.

                        theorem Semantics.Questions.MentionSome.choice_implies_mentionSome {W : Type u_1} {E : Type u_2} (msi : MentionSomeInterrogative W E) (w : W) (y : E) (hy_mem : y msi.whDomain) (hy_sat : msi.abstract w y = true) :
                        (msi.whDomain.any fun (x : E) => msi.abstract w x) = true

                        Choice reading implies mention-some reading.

                        @cite{groenendijk-stokhof-1984}, Section 5.3, p. 538: "The choice reading of (24) implies its mention-some reading. This is correct, to know of a particular pen who has that pen, implies to know a person who has a pen."

                        A choice answer selects a specific satisfier from the wh-domain. Any such witness directly satisfies the existential required by mention-some.

                        For the full two-domain case (wide-scope existential over a separate domain), see wideScope_existential_licenses_mentionSome in ScopeReadings.lean.

                        theorem Semantics.Questions.MentionSome.mentionAll_implies_mentionSome {W : Type u_1} {E : Type u_2} (msi : MentionSomeInterrogative W E) (w : W) (satisfiers : List E) (hAll : satisfiers = List.filter (msi.abstract w) msi.whDomain) (hNonempty : satisfiers.length > 0) :
                        (msi.whDomain.any fun (x : E) => msi.abstract w x) = true

                        Mention-all implies mention-some for non-empty extensions.

                        If you know ALL satisfiers, and there is at least one, you know SOME.

                        Verbs That Block Mention-Some #

                        @cite{groenendijk-stokhof-1984}, Section 5.4: Some verbs BLOCK mention-some readings:

                        These verbs require knowing the COMPLETE answer because they express functional dependencies or relevance that needs full information.

                        In contrast, these verbs ALLOW mention-some:

                        Does a verb allow mention-some readings?

                        @cite{groenendijk-stokhof-1984}, Section 5.4: "depends", "matter", "determine" block mention-some because they require complete functional information.

                        Equations
                        Instances For
                          structure Semantics.Questions.MentionSome.EmbeddedQuestionSentence (W : Type u_1) (E : Type u_2) :
                          Type (max u_1 u_2)

                          A sentence with embedded question and verb.

                          • verb : String

                            The matrix verb

                          • The embedded question

                          • subject : E

                            The matrix subject

                          • mentionSomePossible : Bool

                            Does the sentence have mention-some reading?

                          Instances For

                            Mention-Two / Mention-N #

                            @cite{groenendijk-stokhof-1984}, Section 5.3 (following @cite{belnap-1982}): Cumulative quantification with numerals gives "mention-n" readings:

                            "Where do two unicorns live?"

                            This is ambiguous:

                            1. Mention-some: One place where (at least) two unicorns live
                            2. Mention-two (cumulative): Two places that together house two unicorns (e.g., one unicorn in Paris, one in Rome)
                            3. Choice: Identify the specific two unicorns, then where they live

                            The cumulative reading requires the places to COLLECTIVELY satisfy the numeral, not each place individually.

                            structure Semantics.Questions.MentionSome.MentionNQuestion (W : Type u_1) (E : Type u_2) :
                            Type (max u_1 u_2)

                            A mention-n question with cumulative quantification.

                            "Where do N unicorns live?" - asks for places collectively covering N entities.

                            • n :

                              How many entities must be covered

                            • whDomain : List E

                              The wh-domain (places)

                            • entityDomain : List E

                              The entity domain (unicorns)

                            • relation : WEEBool

                              The relation: lives(w, place, entity)

                            Instances For
                              def Semantics.Questions.MentionSome.collectivelyCovers {W : Type u_1} {E : Type u_2} [DecidableEq E] (mnq : MentionNQuestion W E) (places : List E) (w : W) :

                              Does a set of places collectively cover n entities?

                              For mention-n, we need the UNION of entities at these places to have size ≥ n.

                              Equations
                              • One or more equations did not get rendered due to their size.
                              Instances For
                                def Semantics.Questions.MentionSome.mentionSomeAnswerToMentionN {W : Type u_1} {E : Type u_2} [DecidableEq E] (mnq : MentionNQuestion W E) (place : E) (w : W) :

                                Mention-some answer to mention-n: ONE place with n entities.

                                "Where do two unicorns live?" (mention-some) = a single place with 2+ unicorns

                                Equations
                                Instances For
                                  def Semantics.Questions.MentionSome.cumulativeAnswer {W : Type u_1} {E : Type u_2} [DecidableEq E] (mnq : MentionNQuestion W E) (places : List E) (w : W) :

                                  Cumulative mention-n: multiple places collectively covering n entities.

                                  "Where do two unicorns live?" (cumulative) = places that together have 2 unicorns

                                  Equations
                                  Instances For

                                    Categorematic Quantifier Grounding #

                                    @cite{belnap-1982}'s categorematic principle: the same quantifier words (every, some, most) work identically in declarative and interrogative contexts. The ∃ in I-MS is structurally the same generalized-quantifier application as some_sem in Semantics.Lexical.Determiner.Quantifier: both compute domain.any (λ x => restrictor(x) ∧ scope(x)).

                                    def Semantics.Questions.MentionSome.gqApply {E : Type u_1} (domain : List E) (restrictor scope : EBool) :

                                    GQ application: ∃x ∈ domain. R(x) ∧ S(x). This is the shared computation underlying both declarative some_sem and interrogative I-MS.

                                    Equations
                                    Instances For
                                      theorem Semantics.Questions.MentionSome.applyToProperty_eq_gqApply {W : Type u_1} {E : Type u_2} [DecidableEq E] (msi : MentionSomeInterrogative W E) (Q : GSQuestion WBool) (w : W) :
                                      msi.applyToProperty Q w = gqApply msi.whDomain (msi.abstract w) fun (x : E) => Q (yesNoQuestionFor msi.abstract x)

                                      @cite{belnap-1982}'s categorematic principle: applyToProperty factors as gqApply(whDomain, abstract(w), λx. Q(yn-question(x))) — the same GQ existential used in declarative some_sem.

                                      Partial Answer Taxonomy #

                                      @cite{belnap-1982} distinguishes direct answers (complete, non-redundant) from partial and eliminative answers. Two partial-answer definitions exist in linglib:

                                      isPartialAnswer is strictly stronger: it additionally requires that p fails to overlap with at least one cell (ruling out tautological propositions).

                                      theorem Semantics.Questions.MentionSome.partial_answer_notions_relationship :
                                      have q := QUD.ofProject id; have worlds := [0, 1, 2]; have p := fun (w : Fin 3) => decide (w < 2); Answerhood.isPartialAnswer p q worlds = true partialAnswer p q worlds = true Answerhood.isPartialAnswer (fun (x : Fin 3) => true) q worlds = false partialAnswer (fun (x : Fin 3) => true) q worlds = true

                                      The two partial-answer notions agree on genuine partial answers but diverge on tautologies. Concrete witness on Fin 3 with the identity partition.