SFB 441 - Project B11

Realisation of arguments

The annotation revealed that, despite great differences in grammar and style and in the length of the annotation, all texts showed a similar behaviour with respect to argument realisation: more than 40% of all arguments remained non-realised, see Fig. 1.

The texts differ, however, in how many non-realised arguments have an explicit antecedent in the preceding text. This difference is pragmatically conditioned and genre-sensitive: The TVP has many doctrinal passages which do not refer to specific human actors and thus the corresponding arguments remain unspecified even without having an antecedent.

Fig. 1: Relation between expected and non-realised arguments (total)

database	OTC	%	TVP	%	LLV	%
expected arguments (real complements)	505		1795		1335
realised arguments	269	53	999	56	758	57
empty arguments	198	39	639	36	440	33
empty arguments with antecedent	156	(79)	421	(66)	333	(76)
empty arguments without antecedent	42	(21)	218	(34)	107	(24)
omitted arguments	39	8	157	9	133	10
omitted arguments with antecedent	21	(54)	27	(17)	33	(25)
demonstrative pronouns	12		123		105
demonstrative pronouns with antecedent	8	(67)	95	(77)	98	(93)
demonstrative pronouns without antecedent	4	(33)	28	(23)	7	(7)

other figures
Distribution of arguments (the ten most frequent roles)
Sensitivity of roles for deletion or omission
Congruence patterns between antecedents and anaphoric elements
Reference relations for zero-anaphora (distance-sentitive)

top

Distribution of arguments (the ten most frequent roles)

Fig. 2 shows the ten most common roles, with and without regard of valency. For all roles and abbreviations see our definition of arguments. Somewhat unexpected, the texts show quite a similar (in part even identical) distribution for the first five most frequent roles. Among the next frequent roles, one finds more variation, but still some similarities. The high frequency of location arguments is particularly noteworthy: if valency is disregarded, LCT constitutes the third-most frequent role in all three texts (OTC: 12,28%, TVP: 10,03%, LLV: 13,00%), after eA (20,20%, 18,94%, 22,84%) and P (13,27%, 12,70%, 15,56%). If valency is regarded, it is even the most frequent role (12,28%, 10,07%, 13%) before eA2 (8,32%, 9,81%, 10,07%) and P2 (TVP 8,97%) or R (OTC 7,92%, LLV 9,32%). This reflects the frequent occurrence of motion, position and existence verbs, while the high number of omissions (see Fig. 3) reflects the overall pertinence of locations throughout longer chains of events.

Fig. 2 The ten most frequent roles in the texts

valency-sensitive
	OTC			TVP			LLV
	roles	absolute	%	roles	absolute	%	roles	absolute	%
1	LCT	62	12,28	LCT	180	10,03	LCT	173	13,00
2	eA2	42	8,32	eA2	176	9,81	eA2	134	10,07
3	R	40	7,92	P2	161	8,97	R	124	9,32
4	P2	35	6,93	R	106	5,91	P2	94	7,06
5	CONTdir	31	6,14	CONTdir	103	5,74	eA2+	93	6,99
6	eA2+	27	5,35	eA2+	102	5,68	CONTdir	77	5,79
7	eA3	26	5,15	A1+	76	4,23	eA3	77	5,79
8	P3	24	4,75	U1	63	3,51	P3	66	4,96
9	U1	23	4,55	Ep2	56	3,12	HEAD	59	4,43
10	A1+	22	4,36	eA3	55	3,06	ATR	52	3,91
valency-non-sensitive
	OTC			TVP			LLV
	roles	absolute	%	roles	absolute	%	roles	absolute	%
1	eA	102	20,20	eA	340	18,94	eA	304	22,84
2	P	67	13,27	P	228	12,70	P	207	15,55
3	LCT	62	12,28	LCT	180	10,03	LCT	173	13,00
4	A	42	8,32	U	126	7,02	R	124	9,32
5	R	40	7,92	A	106	5,91	CONTdir	77	5,79
6	U	39	7,72	R	106	5,91	A	63	4,73
7	CONTdir	31	6,14	CONTdir	103	5,74	EXST	61	4,58
8	Ep	13	2,57	EXST	84	4,68	HEAD	59	4,43
9	CONTind	12	2,38	Ep	57	3,18	ATR	52	3,91
10	AFCp	11	2,18	AFCp	54	3,01	U	47	3,53

other figures
Realisation of arguments
Sensitivity of roles for deletion or omission
Congruence patterns between antecedents and anaphoric elements
Reference relations for zero-anaphora (distance-sentitive)

top

Sensitivity of roles for deletion or omission

Fig. 3 gives an overview, which roles are most sensitive to non-realisation. These are clearly the most salient roles, above all those related to humanity, that is semantic agents [+ctr] (A, eA), recipients (R), and ceptors [–ctr] (receptors and perceptors, Ep2). The most prototypical agent (eA2) of a prototypical transitive transformative [+ctr] action is realised only in about 25 to 36% of all instances, in the case of higher valency (eA2+, eA3), the percentage is even lower. Self-controlled movers and similar [+ctr] non-ergative arguments (A1+, A2) are also likely to be deleted. recipients, and ceptors are treated somewhat different in the three texts.

Patients are much less likely to be non-realised. However, whenever a chain of events concerns one and the same patient, it has to be obligatorily deleted in all clauses following ist explicit mentioning. This yields a fairly high number of symmetric P < P relations (deleted Ps following a P-antecedent). Most arguments are obligatory arguments, only the locational arguments (LCT, SRC) and the addressee (R) are frequently omissible. Fig. 2 also shows that the more salient arguments occur with a higher frequency than less salient arguments, particularly if one disregards the valency distinctions: patients and recipients are more frequent than subjects of low transitivity. The high frequency of location arguments is noteworthy. For all roles and abbreviations see our definition of arguments.

(If diacritics are not properly displayed, please select the UTF-8 charset on your browser.)

Fig. 3: Sensitivity of roles for deletion or omission (exp. = expected, real. = realised, del. = deleted, om. = omitted)

peripheral non-subject roles
document	OTC					TVP					LLV
realisation	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.
ATR	1	1	100,00	0	–	48	48	100,00	0	–	52*	52	100,00	0	–
RST	0	–	–	–	–	13	13	100,00	0	–	7	6	85,71	1	–
PST	0	–	–	–	–	14	11	78,57	2	1	4	4	100,00	0	–
bArg	6	4	66,67	2	–	26	25	96,15	1	–	3	3	100,00	0	–
PARA	4	4	100,00	0	–	10	10	100,00	0	–	0	–	–	–	–
AFCp2	11	10	90,01	1	–	54	38	70,37	16	–	22	21	95,45	1	–
AFCe2	0	–	–	–	–	10	6	60,00	4	–	0	–	–	–	–
PATH	5	3	60,00	2	–	1	1	100,00	0	–	1	1	100,00	0	–
CONTdir	31	26	83,87	4	1	103	96	93,20	7	–	83	65	78,31	18	–
CONTind	12	9	75,00	3	–	12	12	100,00	0	–	2	2	100,00	0	–
CTC	8	7	87,50	1	–	12	11	91,67	1	–	1	0	0,00	1	–
INSTR	6	5	83,33	1	–	20	18	90,00	2	–	1	1	100,0	0	–
SRC	2	2	100,00	0	0	20	17	85,00	0	3	16	11	68,75	0	5
CREL	0	–	–	–	–	6	6	100,00	0	–	0	–	–	–	–
demA	6	6	100,00	0	0	2	2	100,0	0	–	0	–	–	–	–
LCT	63	51	80,95	3	9	180	119	66,11	11	50	173	87	50,29	15	71
peripheral subject roles
document	OTC					TVP					LLV
realisation	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.
HEAD	1	1	100,00	0	–	48	27	56,25	21	–	59	15	25,42	44*	–
EXST1	0	–	–	–	–	3	3	100,00	0	–	11	11	100,00	0	–
EXST1+	3	3	100,00	0	–	40	34	85,00	6	–	50	40	80,00	10	–
primary non-subject roles
document	OTC					TVP					LLV
realisation	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.
TAR	7	4	57,14	3	–	18	9	50,00	8	1	27	23	85,19	4	–
R	40	7	17,50	8	25	106	25	23,58	15	66	124	61	49,19	19	44
P2	35	27	77,14	8	–	161	129	80,12	32	–	94	88	93,62	6	–
P2+	7	4	57,14	3	–	24	17	70,83	7	–	47	27	57,45	20	–
P3	24	23	95,83	1	–	39	25	64,10	14	–	66	57	86,36	9	–
P3+	0	–	–	–	–	3	1	33,33	2	–	0	–	–	–	–
P4	1	1	100,00	0	–	1	0	0,00	1	–	0	–	–	–	–
primary subject roles [–ctr]
document	OTC					TVP					LLV
realisation	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.
EXST2	6	6	100,00	0	–	37	37	100,00	0	–	0	–	–	–	–
EXST2+	0	–	–	–	–	3	2	66,67	1	–	0	–	–	–	–
EXST3	0	–	–	–	–	1	1	100,00	0	–	0	–	–	–	–
U1	23	18	78,26	5	–	63	44	69,84	19	–	20	17	85,00	3	–
U1+	3	0	0,00	3	–	9	6	66,67	3	–	7	7	100,00	0	–
U2	13	7	53,85	6	–	54	23	42,59	29	2	18	11	61,11	7	–
U3	0	–	–	–	–	0	–	–	–	–	2	1	50,00	1	–
Ee2	0	–	–	–	–	10	6	60,00	4	–	0	–	–	–	–
Eg2	0	–	–	–	–	5	5	100,00	0	–	0	–	–	–	–
Ee3	0	–	–	–	–	0	–	–	–	–	1	1	100,00	0	–
Ep2	14	1	7,14	13	–	57	19	33,33	38	–	24	15	62,50	9	–
Ep2+	0	–	–	–	–	1	1	100,00	0	–	0	–	–	–	–
POSS	6	2	33,33	4	–	36	20	55,56	16	–	–	–	–	–	–
CU1	0	–	–	–	–	10	8	80,00	2	–	0	–	–	–	–
CU1+	0	–	–	–	–	1	1	100,00	0	–	0	–	–	–	–
CU2	0	–	–	–	–	3	3	100,00	0	–	0	–	–	–	–
CEXST1+	0	–	–	–	–	1	1	100,00	0	–	0	–	–	–	–
CA1+	0	–	–	–	–	1	1	100,00	0	–	0	–	–	–	–
CeA2	0	–	–	–	–	1	0	0,00	1	–	0	–	–	–	–
ME1	0	–	–	–	–	0	–	–	–	–	1	1	100,00	0	–
ME1+	1	1	100,00	0	–	0	–	–	–	–	0	–	–	–	–
ME2	2	0	0,00	2	–	0	–	–	–	–	0	–	–	–	–
eME2	1	0	0,00	1	–	0	–	–	–	–	0	–	–	–	–
E₀	0	–	–	–	–	0	–	–	–	–	13	12	92,31	1	–
primary subject roles [+ctr]
document	OTC					TVP					LLV
realisation	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.
A1	2	1	50,00	1	–	2	0	0,00	2	–	1	1	100,00	0	–
A1+	22	5	22,73	17	–	76	25	32,89	51	–	51	25	49,02	26	–
A2	18	7	38,89	11	–	27	6	22,22	21	–	9	6	66,67	3	–
A2+	0	–	–	–	–	1	1	100,00	0	–	0	–	–	–	–
A3	0	–	–	–	–	0	–	–	–	–	2	1	50,00	1	–
A2/eA2rfl	0	–	–	–	–	2	2	100,00	0	–	3	0	0,00	3	–
eA1	0	–	–	–	–	2	0	0,00	2	–	10	0	0,00	10	–
eA1rcp	0	–	–	–	–	0	–	–	–	–	1	1	100,00	0	–
eA2	44	11	25,00	33	–	184	48	26,09	136	–	137	49	35,77	88	–
eA2+	27	3	11,11	24	–	108	21	19,44	87	–	93	17	18,28	76	–
eA3	27	6	22,22	21	–	56	7	12,50	49	–	77	14	18,18	63	–
eA3+	1	0	0,00	1	–	3	0	0,00	3	–	0	–	–	–	–
eA4	6	0	0,00	6	–	1	0	0,00	1	–	0	–	–	–	–
CIN	0	–	–	–	–	18	4	22,22	14	–	0	–	–	–	–
the most salient and/or frequent roles – disregarding valency
document	OTC					TVP					LLV
realisation	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.	exp.	real.	% real.	del.	om.
TAR	7	4	57,14	3	–	18	9	50,00	8	1	27	23	85,19	4	–
U (1-2)	39	25	64,10	14	–	125	73	58,40	51	2	47	36	76,60	11	–
A (1-2+)	42	13	30,95	29	–	106	32	30,19	74	–	63	33	52,38	30	–
R	40	7	17,50	8	25	106	25	23,58	15	66	124	61	49,19	19	44
LCT	60	48	80,00	3	9	180	119	66,11	11	50	173	87	50,29	15	71
P (2-4)	67	55	82,09	12	–	228	172	75,44	56	–	207	172	83,09	35	–
eA (2-4)	105	20	19,05	85	–	354	76	21,47	278	–	307	80	26,06	227	–

*The occurrence of so many HEAD and ATR pairs in the LLV, and even more the high number of deletions of HEAD arguments is an artefact, due to a repetitious song.

For all roles and abbreviations see our definition of arguments.

other figures
Realisation of arguments
Distribution of arguments (the ten most frequent roles)
Congruence patterns between antecedents and anaphoric elements
Reference relations for zero-anaphora (distance-sentitive)

top

Congruence patterns between anaphora and their antecedents

We have evaluated the annotations for the relation between empty arguments and their antecedents with respect to their roles. While we believe that valency distinctions play a crucial role and should not be neglected (or not in all cases: they seem to be more important for less prototypical ‘transitive’ roles than for those of the upper section of the transitive hierarchy), it is also clear that the combinatory potential is growing exponentially, and in the case of the TVP (although not in the case of the LLV) we got, in fact too much ‘noise’, that is too many combinations with only single or low numbers of occurrences. We, therefore run the evaluation a second time, taking out all valency distinctions, except that between eA1 (single argument in the ergative) and all other ergative arguments (eA2, eA2+. etc.), see also the last section of Fig. 3 above. We kept all other semantically ([±control], event-type) or morphologically (case marking) relevant distinctions.

Since practically (almost) all possible combinations concerning the roles of an anaphoric element and its antecedent are attested, only weak preferences (around 75%) for congruence with respect to either control or subjecthood can be observed. Control or subjecthood congruence means that either both antecedent and anaphora display the features or both lack them. The latter, i.e. congruence with respect to [–control] and [–subjecthood] is clearly the preferred pattern for pronominal anaphora. But the opposite is not true for zero-anaphora, even though both types of anaphora are functionally distributed.

We have not yet annotated enough data to draw robust conclusions, but it seems that the Old Tibetan text prefers [±control]-congruence over [±subjecthood]-congruence, while the Classical Tibetan text shows the opposite preference. In both cases, congruence for [±control] means that arguments lacking control, e.g. Ps and Us (and possibly Ep.s) are much more likely to be paired than (subject) arguments displaying control (As and eAs). This preference appears to be even higher in the classical text. The chance that subjects are paired is in both cases only slightly higher than 50%.

Interestingly enough, the preferences for subjecthood and control congruence as well as a combination of both are stronger in the Ladakhi text. Even more striking is the preference for subjecthood among the zero-anaphora, going along with a non-preference for negative or positive control values. This difference may reflect the differences between written and oral literature as much as an ongoing language change.

Fig. 4: Congruence patterns between anaphora and their antecedents

all anaphora
	OTC		TVP		LLV
all instances	215	%	615	%	637	%
[±subjecthood] congruence	106	49,30	464	75,45	513	80,53
[+subjecthood]	102	47,44	275	44,72	347	54,47
[±control] congruence	168	78,14	427	69,43	529	83,05
[+control] (& [+subjecthood])*	72	33,49	129	20,98	227	35,64
[±subj] & [±ctr] congruence	87	40,47	368	59,84	455	71,43
role congruence valency sensitive	37	17,21	91	14,80	149	23,39
role congruence valency non-sensitive	53	24,65	129	20,98	256	40,19
case congruence	104	48,37	216	35,12	331	51,96
zero-anaphora
	OTC		TVP		LLV
all instances	171	%	444	%	398	%
[±subjecthood] congruence	102	59,65	331	74,55	346	86,93
[+subjecthood]	102	59,65	258	58,11	288	72,36
[±control] congruence	133	77,78	278	62,61	333	83,67
[+control] (& [+subjecthood])*	72	42,11	126	28,38	213	53,52
[±subj] & [±ctr] congruence	83	48,54	241	54,28	306	76,88
role congruence valency sensitive	32	18,71	74	16,67	77	19,35
role congruence valency non-sensitive	48	28,07	112	25,23	174	43,72
case congruence	97	56,73	187	42,12	205	51,51
pronominal anaphora
	OTC		TVP		LLV
all instances	44	%	171	%	239	%
[±subjecthood] congruence	4	9,09	133	77,78	167	69,87
[+subjecthood]	0	0,00	17	9,94	59	24,69
[±control] congruence	35	79,55	149	87,13	196	82,01
[+control] (& [+subjecthood])*	0	0,00	3	1,75	14	5,86
[±subj] & [±ctr] congruence	4	9,09	127	74,27	149	62,34
role congruence valency sensitive	5	11,36	17	9,94	72	30,13
role congruence valency non-sensitive	5	11,36	17	9,94	82	34,31
case congruence	7	15,91	29	16,96	126	52,72

*The feature [+ctr] already includes subjecthood (but not vice versa).

other figures
Realisation of arguments
Distribution of arguments (the ten most frequent roles)
Sensitivity of roles for deletion or omission
Reference relations for zero-anaphora (distance-sentitive)

top

Reference relations for zero-anaphora (distance-sensitive)

While there is no significant general preference for congruence patterns or particular roles, some combinations do occur more frequently than others, among them clearly agent-centred roles [+ctr]. The most common among these is the congruent combination of two ergative agents, or combinations between ergative and absolutive agents. The overall frequencies depend, however, on the number of clauses that intervene. While it is typically actor related combinations that hold over long distances, one may find other roles more frequently over short distances.

This becomes most strikingly apparent in the case of the LLV, where congruent patient-patient relations turn out to be clearly the most frequent relations over a distance of one clause only. Only if all agent-agent relations are summed up, whether they show case congruence or not, they outnumber the patient-patient relations.

This is, in a way, quite surprising, since patients occur by about one third less frequently than ergative agents (as ergative agents also occur with patient-less directional activity verbs), not to speak of all agents together, and patients are also much less frequently omitted (see fig. 3 above). This latter outcome, of course, reflects the simple fact that patients are not commonly acted upon in a multiple way. From this it follows that if such multiple acting upon identical patient does occur, patients have to be obligatorily deleted in all following sequences, except, perhaps, if due to growing distance, the reference becomes ambiguous. (Yet all beginners in Tibetan studies suffer considerably because ambiguities are more often than not – not resolved.) Quite naturally, identical patients might be part of a clause immediately following their first mention (an ideal szenario would be the preparation of something for further action, e.g. peeling, washing, or cutting a fruit for consumtion). Agents, by contrast, may be deleted over comparatively long distances, and long distances, again allow more possibilities of combinations. For these reasons the likelyhood of agent-agent relations decreases somewhat over short distances.

The LLV also shows an extreme case, where agent-agent relations were outnumbered by a peripheral relation over a distance of two clauses: the deleted head of a predication verb referring back to a PATIENT. The high number of head-arguments in this text and their deletion is not very significant, since it is the result of endless repetitions in songs or enumerations of treasuries.

That the high frequency of patient-patient relations over short distances is not a genre-relevant artefact in the LLV, is also shown by the results of the two other texts, even if the frequencies are clearly lower than those of congruent (ergative) agent-relations.

In order to make the results more comparable, and to make the more significant facts apparent, we have reduced all valency distinctions, with the exception of eA1-arguments, that is, ergative marked sole arguments. The following figures, representing reference relations for distances of one to five clauses, have been further cleared off all marginal non-congruent role-relations (represented only in light grey). We will, however, indicate congruent relations of peripheral roles. This must be enough to show that, in principle, every combination is possible.

Key to figs. 5a - 5c:
The embedded circles represent the clause distances, the outer-most circle a distance of 1, the inner-most a distance of 5. The percentages indicated are based on the total number of reference relations over a given distance. Fields with dark grey shade represent references to non-arguments. These may be references to extracted NPs, as in the case of expeditive procedures or nominal introduction of speeches or references to broader situations (clauses, sentences, quotes). Light grey fields represent non-congruent peripheral roles. Monochrome colours are used for role-congruence. In the case of non-congruence, the inner coulour represents the anaphoric element, the outer the antecedent. The colours themselves reflect case marking and, to a limited extend also differences in roles with identical case marking: dark blue represents ergative agent-marking (light blue in the case of single arguments: the case of 'having a look'), the yellowish colours represent absolutive marking (golden-orange for As, bright yellow for Ps, and light yellow for all others), green colours represent locational marking. The typologically non-predicted reference relation between patients and ergative agents (neither case nor subjecthood congruence) is particularly highlighted. It occurs – with low frequency – in all three texts.

Fig. 5 Synopsis

OTC (Fig. 5a)	TVP (Fig. 5b)	LLV (Fig. 5c)

Fig. 5a: Reference relations (zero-anaphora) in OTC
OTC
reference relations

Fig. 5b: Reference relations (zero-anaphora) in TVP
TVP
reference relations

Fig. 5a: Reference relations (zero-anaphora) in LLV
LLV reference relations

Cross clausal cross references had been predicted by Cooreman, A., Fox, B. and Givón, Talmy (1984) as falling into three types:

Neutral reference (case and subjecthood are congruent): eA < eA; P < P
Accusative reference (congruence of subjecthood): eA < S; S < eA
Ergative reference (congruence of case, Abs): P < S; S < P

What was not predicted, is the total non-congruent reference (neither case nor subjecthood): P < eA; eA < P.

Ladakhi and Tibetan do not show any syntactical restriction (or pivot) for deletion of arguments under clause chaining, although one can see a certain preference for congruence of the feature [±control] and neutral and “accusative” references can be observed.

Cooreman, A., Fox, B. and Givón, Talmy. 1984. „The discourse definition of ergativity“, Studies in Language 8: 1–34

Realisation of arguments
Distribution of arguments (the ten most frequent roles)
Sensitivity of roles for deletion or omission
Congruence patterns between antecedents and anaphoric elements

top

Layout: Christoph Singer. Responsible for the content: B. Zeisler. Last modified: 08.04.2009

Project B11:
Semantic roles, case relations, and cross-clausal reference in Tibetan

Statistical evaluation

Project B11: Semantic roles, case relations, and cross-clausal reference in Tibetan

Statistical evaluation

Project B11:
Semantic roles, case relations, and cross-clausal reference in Tibetan