Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
The Radial Category as an Emergent Structure
Hanne M. Eckhoff, Laura A. Janda, and Tore Nesset
Overview
PART ONE: How Do We Know the Structure of a Radial Category?
PART TWO: A Radial Category Emerges from our Data
PART THREE: Why LinguisJc Categories Should be CogniJvely RealisJc
1. APART SP (38) raz-‐pilit’ ‘saw apart’ NP (22) raz-‐gryzt’ ‘gnaw apart’
6. EXCITEMENT SP (29) raz-‐kalit’ ‘make red-‐hot’ NP (16) raz-‐gorjačit’ ‘heat up, irritate’
4. SWELL SP (3) raz-‐dut’ ‘inMlate’ NP (9) raz-‐puxnut’ ‘swell’
3. SPREAD SP (30) raz-‐katat’ ‘roll out’ NP (17) raz-‐vetvit’sja ‘branch out’
2. CRUSH SP (7) raz-‐toptat’ ‘trample’ NP (5) raz-‐davit’ ‘crush’
7. UN-‐ SP (38) raz-‐gruzit’ ‘unload’
5. SOFTEN / DISSOLVE SP (7) raz-‐tvorit’sja ‘dissolve’ NP (6) raz-‐tajat’ ‘melt’
Russian RAZ-‐: a Radial Category
PART ONE: How Do We Know the Structure of a Radial Category?
• A radial category is made up of: – nodes, one (or more) of which is a
protototype – arrangement of nodes
• How do we know what the nodes are? – Nodes can exist on various levels of
abstracJon • How do we idenJfy the prototype?
– Various criteria can point to different candidates
• How do we find the arrangement?
• Linguists have oSen relied on their intui?on (aka experJse, logic) to idenJfy the nodes, prototype, and arrangment of a radial category
• There are some criteria (Taylor 1995, Lewandowska-‐Tomaszczyk 2007), but is this enough?
In our study, the nodes, prototype, and arrangement emerge from the data...
... but they also need our help!
PART TWO: Our Data The verb BYTI ‘be’ in Old Church Slavonic
• Old Church Slavonic is the first literary language of the Slavs, preserved in a collecJon of texts from the 9th-‐11th centuries CE
• Our data is 2,428 a^estaJons of the verb BYTI ‘be’ extracted from the PROIEL corpus h^p://foni.uio.no:3000
• All our data is publicly available at TROLLing: hdl:10037.1/10074
Controversy: Is BYTI ‘be’ one verb or two?
Morphological subparadigm
es-‐forms bū-‐forms
“duplicate” forms for es-‐group and bū-‐group
present jestъ bǫdetъ
imperfect běaše *bǫdeaše
aorist bě by(stъ)
present parJciple sǫšt-‐ bǫdǫšt-‐
unique forms subjuncJve bi —
“shared” forms (bū) past parJciple byvъ
imperaJve bǫdi
infiniJve by8
l-‐form bylъ
Two hypotheses
H2: BYTI ‘be’ is a pair of verbs This is a null hypothesis ∴ We do NOT expect a single coherent radial category
We test the hypotheses by means of: Construc?onal profiling -‐-‐ relaJve frequency of construcJons Gramma?cal profiling -‐-‐ relaJve frequency of grammaJcal forms
H1: BYTI ‘be’ is a single verb But it is clearly complex, polysemous ∴ We expect a single coherent radial category
Two hypotheses
H2: BYTI ‘be’ is a pair of verbs This is a null hypothesis ∴ We do NOT expect a single coherent radial category
We test the hypotheses by means of: Construc?onal profiling -‐-‐ relaJve frequency of construcJons Gramma?cal profiling -‐-‐ relaJve frequency of grammaJcal forms
H1: BYTI ‘be’ is a single verb But it is clearly complex, polysemous ∴ We expect a single coherent radial category
The en?re radial category of BYTI
emerges objec?vely from the data
Two steps: nodes and arrangement
STEP ONE: ConstrucJonal profile of BYTI – RelaJve frequency distribuJon of construcJons that BYTI appears in
– ConstrucJons ≈ meanings of BYTI – Emerges from corpus annotaJon data – This gives us the nodes of the radial category
STEP TWO: GrammaJcal profiles of the BYTI construcJons – RelaJve frequency distribuJon of es-‐ and bū-‐forms for each construcJon
– Emerges from corpus annotaJon data – This gives us the arrangement of the radial category
Construc?onal profiling of BYTI
• Aggregated informaJon from annota?on in PROIEL corpus, including: – syntacJc funcJon (auxiliary, matrix verb, dependent verb) – argument structure (oblique arguments, predicaJve complements & adverbials, part of speech and case marking of arguments)
• Hand sor?ng to separate personal from impersonal and group construcJons according to similar pa^erns
• Yields ten construc?ons ≈ meanings of BYTI
Ten constructions ≈ meanings of BYTI (first five)
aux: auxiliary in perfect tense pravě sǫdilъ esi ‘you have judged rightly’ (Mar. Luke 7:43)
auxiliaroid: used with parJciples pisano estъ ‘it is wri^en’ (Mar. Ma^. 21:13)
copula: vy este solь zemi ‘you are the salt of the earth’ (Zogr. Ma^. 5:13)
copular benefac?ve: with a daJve nominal uněe estъ vamъ da azъ idǫ ‘(it) is be^er for you that I should go’ (Mar.
John 16:7)
copular impersonal: lacks referenJal subject bě že noštъ ‘(it) was night’ (Mar. John 13:30)
Ten constructions ≈ meanings of BYTI (remaining five)
copular possessive: with daJve nominal possessor kako emu estъ synъ ‘how can he be his son?’ (Mar. Luke 20:44)
existen?al: with referenJal subject vъ domu otca moego obitěli mъnogy sǫtь ‘in my father's house there are many rooms’ (Mar. John 14:2)
existen?al impersonal: lacks referenJal subject i bystъ egda sěaše ‘and (it) happened as he was sowing’ (Mar. Mark 4:4)
existen?al possessive: with a daJve nominal ne bě ima čęda ‘they had no child’ (Mar. Luke 1:7)
posi?on/goal: indicaJng posiJon, source, or goal otcъ moi iže estъ na nebesьxъ ‘my father who is in heaven’ (Mar. Ma^. 16:17)
���
�����������
����
������� � �����
��������
������
��������� ����
���� �����
���� �������� ������
���� ������
��� ����
�������
�����
�
���
���
���
���
The construc?onal profile of BYTI
���
�����������
����
������� � �����
��������
������
��������� ����
���� �����
���� �������� ������
���� ������
��� ����
�������
�����
�
���
���
���
���
The construc?onal profile of BYTI
These nodes
emerge from the data
���
�����������
����
������� � �����
��������
������
��������� ����
���� �����
���� �������� ������
���� ������
��� ����
�������
�����
�
���
���
���
���
The construc?onal profile of BYTI
Do these nodes
form a radial category?
�������
������
�����
�����������
���
�������
�����
�����������
����������
����� ���
�����
�
��
��
��
��
The gramma?cal profile of BYTI in the auxiliary construcJon
es-‐forms bū-‐forms shared forms
Gramma?cal profiles: arrangement of the nodes
• Ten construcJons (≈ meanings ≈ nodes) have ten different grammaJcal profiles (= distribuJons of es-‐ and bū-‐forms of BYTI)
• A correspondence analysis measures the distances between the construcJons in terms of the grammaJcal profiles
• This gives us the arrangement of the nodes • IF the arrangement yields a coherent radial category and
there is overlap in use of forms, BYTI is probably one verb (H1)
• IF the arrangement does not yield a single radial category and there is specializa?on and contrast in use of forms, BYTI is probably two verbs (H2)
!
0 1 2 3 4
−0.5
0.0
0.5
1.0
1.5
Factor 1 (56.1 %)
Facto
r 2 (
21.3
%)
aux
auxiliaroid
copulacopular_benefactive
copular_impersonal
copular_possessive
existential
existential_impersonal
existential_possessiveposition_goal
1
2
34
5
Correspondence Analysis of the GrammaJcal Profiles of the Ten BYTI ConstrucJons: yields 5 nodes and radial category arrangement
(Auxiliary construcJon is actually much more peripheral due to subjuncJve,
which is removed in this plot)
Clustering of BYTI constructions and forms cluster construc?on es-‐ forms percent bū-‐ forms percent shared
forms percent
Cluster 1 auxiliary 203 95.8% 9 4.2% 0 0%
Cluster 2 existenJal impersonal
0 0% 47 100% 0 0%
Cluster 3 copula 746 81.0% 121 13.1% 54 5.9%
copular possessive
22 84.6% 4 15.4% 0 0%
copular benefacJve
28 59.6% 15 31.9% 4 8.5%
Cluster 4 existenJal 124 42.8% 147 50.7% 19 6.6%
existenJal possessive
22 42.3% 24 46.2% 6 11.5%
auxiliaroid 201 55.7% 134 37.1% 26 7.2%
posiJon/goal 326 76.3% 77 18.0% 24 5.6%
Cluster 5 copular impersonal
22
48.9% 23 51.1% 0 0%
Clustering of BYTI constructions and forms cluster construc?on es-‐ forms percent bū-‐ forms percent shared
forms percent
Cluster 1 auxiliary 203 95.8% 9 4.2% 0 0%
Cluster 2 existenJal impersonal
0 0% 47 100% 0 0%
Cluster 3 copula 746 81.0% 121 13.1% 54 5.9%
copular possessive
22 84.6% 4 15.4% 0 0%
copular benefacJve
28 59.6% 15 31.9% 4 8.5%
Cluster 4 existenJal 124 42.8% 147 50.7% 19 6.6%
existenJal possessive
22 42.3% 24 46.2% 6 11.5%
auxiliaroid 201 55.7% 134 37.1% 26 7.2%
posiJon/goal 326 76.3% 77 18.0% 24 5.6%
Cluster 5 copular impersonal
22
48.9% 23 51.1% 0 0%
Two central clusters of construc?ons
Clustering of BYTI constructions and forms cluster construc?on es-‐ forms percent bū-‐ forms percent shared
forms percent
Cluster 1 auxiliary 203 95.8% 9 4.2% 0 0%
Cluster 2 existenJal impersonal
0 0% 47 100% 0 0%
Cluster 3 copula 746 81.0% 121 13.1% 54 5.9%
copular possessive
22 84.6% 4 15.4% 0 0%
copular benefacJve
28 59.6% 15 31.9% 4 8.5%
Cluster 4 existenJal 124 42.8% 147 50.7% 19 6.6%
existenJal possessive
22 42.3% 24 46.2% 6 11.5%
auxiliaroid 201 55.7% 134 37.1% 26 7.2%
posiJon/goal 326 76.3% 77 18.0% 24 5.6%
Cluster 5 copular impersonal
22
48.9% 23 51.1% 0 0%
Three peripheral construc?ons
Clustering of BYTI constructions and forms cluster construc?on es-‐ forms percent bū-‐ forms percent shared
forms percent
Cluster 1 auxiliary 203 95.8% 9 4.2% 0 0%
Cluster 2 existenJal impersonal
0 0% 47 100% 0 0%
Cluster 3 copula 746 81.0% 121 13.1% 54 5.9%
copular possessive
22 84.6% 4 15.4% 0 0%
copular benefacJve
28 59.6% 15 31.9% 4 8.5%
Cluster 4 existenJal 124 42.8% 147 50.7% 19 6.6%
existenJal possessive
22 42.3% 24 46.2% 6 11.5%
auxiliaroid 201 55.7% 134 37.1% 26 7.2%
posiJon/goal 326 76.3% 77 18.0% 24 5.6%
Cluster 5 copular impersonal
22
48.9% 23 51.1% 0 0%
Aside from two of the peripheral construc?ons, es-‐ and bū-‐forms overlap
and do not contrast
Clustering of BYTI constructions and forms cluster construc?on es-‐ forms percent bū-‐ forms percent shared
forms percent
Cluster 1 auxiliary 203 95.8% 9 4.2% 0 0%
Cluster 2 existenJal impersonal
0 0% 47 100% 0 0%
Cluster 3 copula 746 81.0% 121 13.1% 54 5.9%
copular possessive
22 84.6% 4 15.4% 0 0%
copular benefacJve
28 59.6% 15 31.9% 4 8.5%
Cluster 4 existenJal 124 42.8% 147 50.7% 19 6.6%
existenJal possessive
22 42.3% 24 46.2% 6 11.5%
auxiliaroid 201 55.7% 134 37.1% 26 7.2%
posiJon/goal 326 76.3% 77 18.0% 24 5.6%
Cluster 5 copular impersonal
22
48.9% 23 51.1% 0 0%
Auxiliarioid is very different from
auxiliary
Clustering of BYTI constructions and forms cluster construc?on es-‐ forms percent bū-‐ forms percent shared
forms percent
Cluster 1 auxiliary 203 95.8% 9 4.2% 0 0%
Cluster 2 existenJal impersonal
0 0% 47 100% 0 0%
Cluster 3 copula 746 81.0% 121 13.1% 54 5.9%
copular possessive
22 84.6% 4 15.4% 0 0%
copular benefacJve
28 59.6% 15 31.9% 4 8.5%
Cluster 4 existenJal 124 42.8% 147 50.7% 19 6.6%
existenJal possessive
22 42.3% 24 46.2% 6 11.5%
auxiliaroid 201 55.7% 134 37.1% 26 7.2%
posiJon/goal 326 76.3% 77 18.0% 24 5.6%
Cluster 5 copular impersonal
22
48.9% 23 51.1% 0 0%
Impersonals very different from personals
Outcome for BYTI
• Our correspondence analysis shows exactly the structure of relationships we would expect to obtain for a polysemous verb like BYTI
• In other words, we get a coherent radial category
• Our findings support H1 that BYTI is a single, polysemous verb
PART THREE: Why Linguis?c Categories Should Be Cogni?vely Realis?c
• A realisJc radial category network structure can be obtained objecJvely
• However, a great deal of intuiJon/experJse is needed in order to – ensure that the input is properly annotated and – properly interpret the data
• Only cogni?vely realis?c categories make it possible to join data analysis to linguis?c insights