An encoder relabelled by a symmetry of the source has the encoder's complexity.
Docs / Studies / Studies.ZaslavskyMaldonadoCulbertson2021
Let’s Talk (Efficiently) about Us: Person Systems Achieve Near-Optimal Compression
Zaslavsky, N., Maldonado, M. & Culbertson, J. (2021). In Proceedings of the 43rd Annual Meeting of the Cognitive Science Society (pp. 938–944)
Zaslavsky, Maldonado and Culbertson apply the Information Bottleneck to personal pronouns. The person space has eleven referents, built from the speaker, the addressee and others, one, exactly two or more of them. A speaker's meaning for a referent spreads over the space in proportion to $\exp(-\gamma d)$, $d$ a weighted Hamming distance between the referents' features, and a pronoun system is an encoder of meanings into forms. Zwicky's generalization, that no language gives the inclusive the form of the second person to the exclusion of the exclusive, is where the paper's models part: weighting the speaker and the addressee alike rates the unattested second-inclusive system as efficient as attested ones, while weighting the speaker sixteen times the addressee, the egocentric model, separates them. With the weights alike, exchanging the speaker and the addressee is a symmetry of the meanings, and under a need distribution that the exchange also preserves an encoder and its exchange have the same complexity and the same accuracy. The second-inclusive system is the exchange of a first-inclusive one, so the unbiased model values the two exactly alike unless the need distribution tells the speaker from the addressee: the separation needs the egocentric weights or an asymmetric need. Harbour's two person features, by contrast, generate the partition of the first system's groups and never that of the second's, Zwicky's tripartition: a hard constraint where the paper finds a soft bias.
Implementation notes
The paper encodes number in two bits that its unbiased model counts as one feature; here the number term of the distance is whether the numbers differ. The precision $\gamma$, which the paper fixes from the speaker's memory capacity, and the need distribution, which it infers from corpora, are parameters.
References
- Zaslavsky et al. 2021, Let’s Talk (Efficiently) about Us: Person Systems Achieve Near-Optimal Compression
- Zaslavsky et al. 2018, Efficient compression in color naming and its evolution formalized
- Harbour 2016, Impossible Persons formalized
Symmetries of the Information Bottleneck
An encoder relabelled by a symmetry of the source that the meanings follow has the encoder's accuracy.
The person space
Fig. 1A: the referents of the person space, by the speaker, the addressee and others they hold and by number, one, exactly two or more than two.
- s : Referent
The speaker.
- so : Referent
The speaker and one other.
- soo : Referent
The speaker and several others.
- sa : Referent
The speaker and the addressee.
- sao : Referent
The speaker, the addressee and others.
- a : Referent
The addressee.
- ao : Referent
The addressee and one other.
- aoo : Referent
The addressee and several others.
- o : Referent
One other.
- oo : Referent
Two others.
- ooo : Referent
More than two others.
Derives DecidableEq, Fintype.
Used by Referent.instMeasurableSpace, Referent.instDiscreteMeasurableSpace, Referent.instNonempty, Referent.speaker, Referent.addressee, Referent.others, Referent.number, Referent.swap, Referent.swap_swap, Referent.swapEquiv, Weights.dist, dist_swap, meanings, meanings_swap, firstInclusive, secondInclusive, objective_secondInclusive, Referent.participants, groupGraph
Built from
Built from
Built from
The referent holds the speaker.
- Referent.s.speaker = true
- Referent.so.speaker = true
- Referent.soo.speaker = true
- Referent.sa.speaker = true
- Referent.sao.speaker = true
- x✝.speaker = false
Built from
Used by Referent.participants
The referent holds the addressee.
- Referent.sa.addressee = true
- Referent.sao.addressee = true
- Referent.a.addressee = true
- Referent.ao.addressee = true
- Referent.aoo.addressee = true
- x✝.addressee = false
Built from
Used by Referent.participants
The referent holds an other.
- Referent.s.others = false
- Referent.sa.others = false
- Referent.a.others = false
- x✝.others = true
Built from
The referent's number: one, exactly two, or more than two.
- Referent.s.number = 0
- Referent.a.number = 0
- Referent.o.number = 0
- Referent.so.number = 1
- Referent.sa.number = 1
- Referent.ao.number = 1
- Referent.oo.number = 1
- Referent.soo.number = 2
- Referent.sao.number = 2
- Referent.aoo.number = 2
- Referent.ooo.number = 2
Built from
Exchanging the speaker and the addressee.
- Referent.s.swap = Referent.a
- Referent.so.swap = Referent.ao
- Referent.soo.swap = Referent.aoo
- Referent.a.swap = Referent.s
- Referent.ao.swap = Referent.so
- Referent.aoo.swap = Referent.soo
- x✝.swap = x✝
Built from
Used by Referent.swap_swap, Referent.swapEquiv, dist_swap, egocentric_dist_swap_ne
The exchange of the speaker and the addressee as a permutation.
- Referent.swapEquiv = Function.Involutive.toPerm Referent.swap Referent.swap_swap
Built from ,
Used by meanings_swap, secondInclusive_eq, objective_secondInclusive
The weights of the speaker, addressee, others and number features.
- speaker : ℝ
The weight of the speaker feature.
- addressee : ℝ
The weight of the addressee feature.
- others : ℝ
The weight of the others feature.
- number : ℝ
The weight of number.
Used by Weights.dist, egocentric, dist_swap, meanings, instIsMarkovKernelReferentMeanings, meanings_swap, objective_secondInclusive
The weighted Hamming distance between two referents' features.
The definition is too large to print here; read it in the source.
Built from ,
Used by dist_swap, egocentric_dist_swap_ne, meanings
The egocentric weights of the paper's egocentric model.
- egocentric = { speaker := 16, addressee := 1, others := 0.1, number := 1 }
Built from
Used by egocentric_dist_swap_ne
Built from , ,
The speaker's meaning of a referent t with precision γ, the distribution
$m_t(u) \propto \exp(-\gamma d(u, t))$.
- meanings w γ = ProbabilityTheory.Kernel.ofWeights fun (t u : Referent) => ENNReal.ofReal (Real.exp (-γ * w.dist u t))
Built from , ,
Used by instIsMarkovKernelReferentMeanings, meanings_swap, objective_secondInclusive
With the speaker and the addressee weighted alike, the meaning of the exchanged referent is the exchanged meaning.
Built from , , ,
Inclusive systems
The first-inclusive system of Spanish in Fig. 1B: a form for each person alone and another for its groups, the inclusive taking the form of the speaker's groups.
- firstInclusive Referent.s = 0
- firstInclusive Referent.so = 1
- firstInclusive Referent.soo = 1
- firstInclusive Referent.sa = 1
- firstInclusive Referent.sao = 1
- firstInclusive Referent.a = 2
- firstInclusive Referent.ao = 3
- firstInclusive Referent.aoo = 3
- firstInclusive Referent.o = 4
- firstInclusive Referent.oo = 5
- firstInclusive Referent.ooo = 5
Built from
Used by secondInclusive_eq, objective_secondInclusive, groupGraph_firstInclusive, groupGraph_mem_image_setting_graph
The unattested second-inclusive system of Fig. 1B, the inclusive taking the form of the addressee's groups.
- secondInclusive Referent.s = 2
- secondInclusive Referent.so = 3
- secondInclusive Referent.soo = 3
- secondInclusive Referent.a = 0
- secondInclusive Referent.ao = 1
- secondInclusive Referent.aoo = 1
- secondInclusive Referent.sa = 1
- secondInclusive Referent.sao = 1
- secondInclusive Referent.o = 4
- secondInclusive Referent.oo = 5
- secondInclusive Referent.ooo = 5
Built from
Used by secondInclusive_eq, objective_secondInclusive, groupGraph_secondInclusive, groupGraph_mem_image_setting_graph
Built from , ,
Weighting the speaker and the addressee alike, under a need distribution that their exchange preserves, the second-inclusive system and the first-inclusive system have the same value of the Information Bottleneck objective at every trade-off.
Built from , , , , ,
Against Harbour's parameters
On the referents with more than one member, the first-inclusive system groups the inclusive with the exclusive and the second-inclusive system groups it with the second person, the person partitions of the standard tripartition and of Zwicky's.
The participants of a referent.
- r.participants = (if r.speaker = true then {SpeechAct.Role.speaker} else ∅) ∪ if r.addressee = true then {SpeechAct.Role.addressee} else ∅
Built from , ,
The partition a pronoun system induces on the participant sets of groups: two are related when groups with those participants share a form.
The definition is too large to print here; read it in the source.
Built from
Used by groupGraph_firstInclusive, groupGraph_secondInclusive, groupGraph_mem_image_setting_graph
Built from ,
Built from ,
The two systems the unbiased model values alike are told apart by Harbour's parameters, which generate the partition of the first-inclusive system's groups and never that of the second-inclusive system's, Zwicky's tripartition.
Built from , ,