Documentation

Linglib.Phenomena.LexicalTypology.Studies.MajidBosterBowerman2008

Majid, Boster & Bowerman (2008) #

@cite{majid-boster-bowerman-2008}

The cross-linguistic categorization of everyday events: A study of cutting and breaking. Cognition 109(2), 235–250.

Core Contributions #

28 typologically diverse languages, 61 video clips depicting "cutting and breaking" events. Correspondence analysis reveals 4 shared dimensions along which languages categorize these events:

Languages share the dimensionality but vary in how many categories they carve and where they place boundaries.

Integration with linglib #

The 4 dimensions project onto existing RootProfile features:

DimensionProjects onto
Dim 1 (predictability)instrumentType × patientRob (derived)
Dim 2 (tearing)resultType == .separationinstrumentType == .hands
Dim 3 (snap/smash)forceDir (bidirectional vs omnidirectional)
Dim 4 (poke hole)specific event type

Bridge theorems connect to LevinClass and MeaningComponents:

Design #

SeparationEvent is a point in the same feature space that RootProfile defines regions over. A verb is compatible with an event iff the event's feature values fall within the verb root's ranges. This captures the many-to-many mapping between events and verbs that varies across languages.

A separation event characterized by physical properties of the action, instrument, and affected object.

This is the stimulus level: each value is a specific point, not a range. Corresponds to a single video clip in the experiment. Verb roots select ranges over these same dimensions (via RootProfile).

Instances For
    Equations
    • One or more equations did not get rendered due to their size.
    Instances For
      Equations
      • One or more equations did not get rendered due to their size.
      Instances For

        Encodings of representative clips from the appendix. Clip numbers follow @cite{majid-boster-bowerman-2008} Appendix (pp. 248–249). We encode a representative subset spanning all 4 dimensions rather than all 61 clips.

        Clip 9: Slice carrot lengthwise into two pieces with knife. High predictability (sharp blade, rigid 1D object).

        Equations
        • One or more equations did not get rendered due to their size.
        Instances For

          Clip 10: Slice carrot crosswise into multiple pieces with knife.

          Equations
          • One or more equations did not get rendered due to their size.
          Instances For

            Clip 24: Cut rope in two with scissors. High predictability (sharp blade, 1D object).

            Equations
            • One or more equations did not get rendered due to their size.
            Instances For

              Clip 32: Cut carrot in half crosswise with single karate chop. Intermediate predictability (hand, but directed blow).

              Equations
              • One or more equations did not get rendered due to their size.
              Instances For

                Clip 1: Tear cloth into two pieces by hand. Tearing event (Dimension 2).

                Equations
                • One or more equations did not get rendered due to their size.
                Instances For

                  Clip 36: Tear cloth about halfway through with two hands.

                  Equations
                  • One or more equations did not get rendered due to their size.
                  Instances For

                    Clip 19: Snap twig with two hands. Snapping event (Dimension 3).

                    Equations
                    • One or more equations did not get rendered due to their size.
                    Instances For

                      Clip 57: Snap carrot with two hands.

                      Equations
                      • One or more equations did not get rendered due to their size.
                      Instances For

                        Clip 39: Smash flower pot with single hammer blow. Smashing event (Dimension 3).

                        Equations
                        • One or more equations did not get rendered due to their size.
                        Instances For

                          Clip 40: Smash plate with single hammer blow.

                          Equations
                          • One or more equations did not get rendered due to their size.
                          Instances For

                            Clip 53: Break stick in two with single downward chisel blow. A chisel is sharp-edged but used ballistically (single blow), giving intermediate predictability — the sharp edge partially constrains the locus of separation but the ballistic delivery reduces control.

                            Equations
                            • One or more equations did not get rendered due to their size.
                            Instances For

                              Clip 45: Poke hole in cloth stretched between two tables with a twig. Distinguished on Dimension 4. A twig is a pointed implement but not a sharp blade — it breaches the material through puncture, not clean cutting.

                              Equations
                              • One or more equations did not get rendered due to their size.
                              Instances For

                                Clip 7: Push chair back from table (reversible separation).

                                Equations
                                • One or more equations did not get rendered due to their size.
                                Instances For

                                  Clip 33: Open a book (reversible separation).

                                  Equations
                                  • One or more equations did not get rendered due to their size.
                                  Instances For

                                    Predictability of the locus of separation (Dimension 1).

                                    The most important dimension cross-linguistically. Events where the agent has precise control over where separation occurs (sharp blade on a yielding surface) are "cutting"; events where the locus is unpredictable (blow from a hammer, snapping by hand) are "breaking".

                                    This is a derived property, not a primitive — it emerges from the interaction of instrument type, object properties, and manner.

                                    Instances For
                                      Equations
                                      • One or more equations did not get rendered due to their size.
                                      Instances For

                                        Compute predictability from event features.

                                        Predictability emerges from the interaction of instrument, force, and manner — not from instrument alone. A sharp blade used with controlled motion yields high predictability, but the same blade used ballistically (e.g., a chisel struck with a single blow) yields only intermediate predictability. The paper emphasizes that Dimension 1 is continuous and "not adequately captured by any single feature."

                                        Equations
                                        • One or more equations did not get rendered due to their size.
                                        Instances For

                                          Break subtypes within the low-predictability cluster (Dimension 3).

                                          Among events with unpredictable separation, languages further distinguish snapping (pressure from both ends breaks a rigid 1D object) from smashing (a blow fragments a rigid 3D object).

                                          Instances For
                                            Equations
                                            • One or more equations did not get rendered due to their size.
                                            Instances For

                                              Classify break subtype for low-predictability events.

                                              Equations
                                              • One or more equations did not get rendered due to their size.
                                              Instances For

                                                Is this a tearing event? (Dimension 2) Tearing = hand separation of a flat flexible object.

                                                Equations
                                                • One or more equations did not get rendered due to their size.
                                                Instances For

                                                  Is this a "cutting and breaking" event (irreversible material destruction) as opposed to a reversible separation? (Dimension 1 of the first correspondence analysis, before restricting to core cut/break events.)

                                                  Equations
                                                  Instances For

                                                    Is this a hole-poking event? (Dimension 4)

                                                    Poking a hole in a flat flexible object — distinguished from cutting and breaking because the object is not separated into pieces. The paper notes this emerged as a distinct cluster in 5/28 languages. Our encoding uses .none instrument for the twig since InstrumentType lacks a .pointed variant; the diagnostic feature is the combination of surface breach + 2D flexible object.

                                                    Equations
                                                    Instances For

                                                      Is a separation event compatible with a verb root's profile?

                                                      An event is compatible iff each of its feature values falls within the root's range for that dimension. Unconstrained dimensions (range = none) accept any value.

                                                      Equations
                                                      • One or more equations did not get rendered due to their size.
                                                      Instances For

                                                        Consistency checks for paired clips. Clips depicting the same event type with different objects should receive the same dimension classifications.

                                                        English cutting-and-breaking verb categories.

                                                        English has 5 basic categories for material destruction events (plus open, take apart for reversible separations): cut, break, tear, snap, smash.

                                                        Instances For
                                                          Equations
                                                          • One or more equations did not get rendered due to their size.
                                                          Instances For

                                                            English verb assignment for core cutting-and-breaking events.

                                                            Equations
                                                            • One or more equations did not get rendered due to their size.
                                                            Instances For

                                                              Yélî Dnye (Papuan isolate, Rossel Island) verb categories.

                                                              @cite{majid-boster-bowerman-2008}: Yélî Dnye speakers used only 3 different verbs for the 61 clips, yet their categorization still correlates with the 4 cross-linguistic dimensions. Demonstrates that even languages with minimal verb inventories respect the shared dimensional structure.

                                                              The paper reports category count (3 verbs) and how they partition the stimulus space but does not list the specific verb forms. We use abstract labels (v1/v2/v3) for the three categories.

                                                              Limitation: The paper reports (p. 242) that the YD verb for tearing was also used for carrot-cutting events (clips 37, 9) that depict separation along the grain. This grouping principle (grain-alignment) is not captured by our predictability-based model. Our SeparationEvent does not encode grain alignment, so yeliDnyeVerb incorrectly assigns along-the-grain cutting to v1 rather than v3. See yeliDnye_grain_limitation.

                                                              Instances For
                                                                Equations
                                                                • One or more equations did not get rendered due to their size.
                                                                Instances For

                                                                  Yélî Dnye verb assignment (approximate, based on reported categorization patterns in the correspondence analysis).

                                                                  This model uses predictability as the primary split, which captures the overall pattern but misses the grain-alignment grouping (see YeliDnyeCBVerb docstring).

                                                                  Equations
                                                                  • One or more equations did not get rendered due to their size.
                                                                  Instances For

                                                                    Tzeltal (Mayan) has the most fine-grained inventory in the sample: speakers used 50+ different verbs for the 61 clips (p. 243). Despite this extreme specificity, Tzeltal's categorization still correlates with the 4 shared dimensions — the dimensions are robust to both very coarse (Yélî Dnye, 3 verbs) and very fine (Tzeltal, 50+ verbs) inventories. We do not model Tzeltal verb assignment because the paper does not provide sufficient data on individual verb-to-clip mappings for 50+ verbs.

                                                                    All three languages agree on the superordinate cut/break boundary (Dimension 1): high-predictability events get a "cutting" verb, low-predictability events get a "breaking" verb. The languages differ in how finely they subdivide the breaking domain.

                                                                    English distinguishes tearing from breaking; Yélî Dnye groups tearing with snapping (both involve hand action).

                                                                    Low-predictability events correspond to Levin's break class: change of state + causation, but NO contact, motion, or instrument specification. The absence of contact/motion/instrumentSpec reflects the fact that break verbs are underspecified for manner — consistent with the unpredictability of the separation locus.

                                                                    The cut/break distinction in MeaningComponents captures the endpoints of Dimension 1: cut specifies manner (instrumentSpec, contact, motion) while break does not.

                                                                    Both cut and break map to accomplishment templates — they share event structure ([ACT CAUSE [BECOME ⟨STATE⟩]]) and differ only in root content. This explains why the cut/break distinction is about manner/predictability (root-level), not about event structure (template-level).

                                                                    Rather than defining inline profiles, we derive them from the actual Fragment entries in Fragments.English.Predicates.Verbal. This ensures that compatibility theorems test the real lexical data.

                                                                    A snapping event IS compatible with break (fracture result, moderate force) — English break can cover snapping events, though snap is more specific.

                                                                    The 4 dimensions are not independent: tearing events (Dim 2) always have low predictability (Dim 1).

                                                                    High-predictability events are never tearing events. Contrapositive of: tearing → low predictability.

                                                                    The superordinate cut/break distinction (Dimension 1) is exhaustive: every event is either high, intermediate, or low predictability.

                                                                    Dimension 3 (snap/smash) is nested within Dimension 1: the snap vs smash distinction only applies to low-predictability events. Events with high or intermediate predictability always have breakSubtype .other. This formalizes the hierarchical structure visible in the correspondence analysis (Fig. 4).

                                                                    The paradigmatic hole-poking event (clip 45) has low predictability — it sits in the "breaking" side of Dimension 1.

                                                                    The key relativity finding: languages share the dimensional structure but place category boundaries at different points. English makes finer distinctions than Yélî Dnye — and crucially, the languages disagree on intermediate events.

                                                                    English and Yélî Dnye DISAGREE on tearing: English gives tear its own category; Yélî Dnye groups it with snapping under a single verb (hand-action separation).

                                                                    English distinguishes smashing from snapping; Yélî Dnye distinguishes them too (different verbs). This is an agreement on Dimension 3.

                                                                    The chisel-blow event (clip 53, intermediate predictability) is a boundary case: it has a sharp instrument but ballistic delivery. English categorizes it as break (the result matters more than the instrument for English), while Yélî Dnye's v1 covers all high-predictability events.

                                                                    Known limitation of the Yélî Dnye model. Our model assigns clip 9 (slice carrot lengthwise) to v1 (cutting), but the paper reports that Yélî Dnye groups this event with tearing (v3) — both involve separation along the grain of the material. This documents the mismatch: the real YD system uses grain-alignment as a grouping principle that our SeparationEvent cannot express.