Towards an Inferential Lexicon of Event Selecting Predicates for French

towards an inferential lexicon of event selecting predicates for french

Ingrid Falk and Fabienne Martin IWCS 2017, September 21 Universität Stuttgart - SFB 732

motivation

this work Inferential lexicon for French I

describes effect of predicates selecting event denoting arguments

I

on event argument

I

in terms of certainty and polarity

He failed toevent selecting predicate resignembedded event resignevent → certain, polarity − (did not happen)

Long-term goal I

Factuality assessment of events in French newspaper texts

I

Cf. [Saurí and Pustejovsky, 2012] for English

Ingrid Falk and Fabienne Martin

Inferences of French ESPs

3 / 38

automatic factuality assessment [Saurí and Pustejovsky, 2009, Saurí and Pustejovsky, 2012]: I

automatically determine certainty and polarity of events.

[Saurí and Pustejovsky, 2012]’s DeFacto: I

computes factuality using 3 lexical resources I I I

polarity particles: not, none, . . . modality particles: may, necessary, . . . event selecting predicats (ESPs): manage to, fail to, . . .

This work: I

build a seed lexicon of event selecting predicates for French

I

capturing the effect on the factuality of embedded events



4 / 38

outline

Motivation Related work Towards a French ESP lexicon Findings Conclusion and Outlook References



5 / 38

related work

English FactBank and French TimeBank [Saurí and Pustejovsky, 2009, Saurí and Pustejovsky, 2012, Bittar, 2010, Bittar et al., 2011] Lexicon from Language and Natural Reasoning (Stanford) [Karttunen, 1971, Nairn et al., 2006]



7 / 38

the english factbank [Saurí and Pustejovsky, 2009, Saurí and Pustejovsky, 2012]: I I

corpus annotated with event factuality TimeBank [Pustejovsky et al., 2005]: events are assigned factuality profiles I I

I

manually [Saurí and Pustejovsky, 2009] automatically [Saurí and Pustejovsky, 2012]

automatic detection based on a lexicon of event selecting predicates

polarity fail I

CT (certain) + −

PR (probable) + −

PS (possible) + −

CT−

PR−

PS−

CT+

PR+

She has failed to leavee the country.



PS+ CT+

→ CT− 8 / 38

the french timebank [Bittar, 2010, Bittar et al., 2011] I I

same principles as the English TimeBank additional markup for linguistic phenomena not yet covered and specific to French

Most relevant for this work I

modal, implicative, factive verbs marked up as events (fully acceptable with perfective and imperfective aspect)

I

account of grammatical tense/aspect system of French eg. imparfait (not grammaticalised in English)

French TimeBank offers I

a sample of French ESPs used in newspaper texts

I

typical embedded events



9 / 38

inferential lexicon I

lexical resource for English from Language and Natural Reasoning group (Stanford) [Nairn et al., 2006]

I

complement-taking verbs (ESPs, ≈ 250)

I

classified w.r.t. polarity of complement clauses (EMB) obtained under positive and negative polarity of ESPs

I

She has failed to leavee the country.

ESP +

→ EMB−

I

She has not failed to leavee the country.

ESP−

→ EMB+

ESP

fail to Ingrid Falk and Fabienne Martin

polarity + −

signature

−

−1|1

+

semantic class 2-way implicative


10 / 38

inferential lexicon: probabilistic signatures

I

introduced by [Karttunen et al., 2016, Karttunen, 2016]

I

reflect the variable strength of the inference

be able → 0.9| − 1 I

under polarity+ ; strong (but defeasible) inference

Ann was able to speak up ; Ann very probably did speak up I

but. . . I I

few examples (≈ 40), not empirically validated (yet).



11 / 38

inferential classification

Polarity of ESP + −

Sample predicate

Signature

factives counterfactives 2-way implicatives

+ − + −

+ − − +

forget that pretend that manage to fail to

1|1 −1|−1 1|−1 −1|1

1-way +implicatives 1-way -implicatives

+ − N N

N N − +

force to prevent to get chance to hesitate to

1|0.5 −1|0.7 0.9|−1 N|1

Neutral

N

N

want to

N|N



12 / 38

towards a french esp lexicon

towards a french ESP lexicon: our experiments Observation Inferential semantic classes → ESP lexicon ESP

fail to

+ −

ESP


− +

signature −1|1

fail to fail to

+ −

semantic class 2-way implicative

embedded event CT + − CT− CT+ CT+ CT−


14 / 38

towards a french ESP lexicon: our experiments Recipe adopted for our French ESP lexicon: 1. start with verbs in inferential classification \ translated to French ESP s

in French TimeBank

2. collect verbal readings as delineated in French lexicons 3. assign probabilistic inferential signatures to readings

Our research questions: I

do inferential signatures vary with outer aspect and animacy of the (deep) subject?

I

do inferential signatures vary with other semantic/syntactic properties?



15 / 38

our data: verbs

ESPs from French TimeBank FTiB [Bittar, 2010, Bittar et al., 2011] \ manual translations of inferential classification by [Nairn et al., 2006] 49 French verbs



16 / 38

our data: readings 1. Extraction of all readings for 49 French verb lemmas from two French valency lexicons:

LVF - [François et al., 2007] refuser 09 Il refuse que Pierre sorte.

Lglex - [Constant and Tolone, 2010] refuser (Table 9) J’ai refusé que Max prenne ma voiture. ≈ 930 readings 2. Manual selection of ESP readings & and suppression of duplicates ≈ 170 readings Ingrid Falk and Fabienne Martin


17 / 38

our data: annotation 170 readings → 3 probabilistic inferential signatures by FM I

with two different aspectual values: perfective PFV and imperfective IMP I

I

±animate (deep) subject for perfective aspect I

value ±1 ±0.9 ±0.8 ±0.7 ±0.6 N

French: inferential profiles vary with outer aspect [Hacquard, 2006] inferential profiles vary with animacy of (deep) subject [Martin and Schäfer, 2012]

strength of inference certain very (un-)likely (un-)likely (not) very possible (not) quite possible no inference


obliger 02 Pierre/cela a obligé Marie à partir. ‘Peter/something force-PAST- PFV.3 SG Mary to go.’ PFV+anim PFV-anim IMP 0.9|N 1|N N|N


18 / 38

aider 01 aider 03 aider 04(s) aider 05 aider V_37M1_12 aimer 03 aimer 04 aimer V_12_8 ajouter 03 apprendre 05 apprendre 03 apprendre 06 assurer 03 assurer 08 assurer V_13_12 assurer V_15_4 autoriser 01 autoriser 03 avouer 01 avouer 02 concéder 01 condamner 01 condamner 03 condamner 09 confirmer 01 confirmer 02 continuer 01 continuer 02 continuer 03 continuer 04 continuer 05 devoir 02 devoir 04 devoir 05 devoir 06 devoir 07 devoir V_1_39 dire 02 dire 03 dire 08 dire 11 échouer 07 échouer V_31R_120 exiger 01 exiger 02 exiger 03 expliquer 03−NOM expliquer 04 expliquer 07(s) garantir 02 garantir 05 garantir 06 hésiter 01 hésiter 05 justifier 02 Justifier 03−ANIM Justifier 03−INANIM justifier 04 justifier 06 justifier 07(ê) justifier V_13_96 justifier V_35R_265 laisser 18 laisser 19 laisser 20(s) laisser 23 laisser V_1_61 laisser V_11_159 lire 06 lire 07 lire 11 lire 12 montrer 04 montrer 05(s) montrer 07 montrer 08 montrer 09 montrer 10 montrer 11 montrer 12 montrer V_9_266 motiver 01 motiver 02 motiver 03 motiver V_10_143 motiver V_11_163 obliger 02 obliger V_4_501 observer 04 observer 06 observer V_32R3_602 penser 04−A−INF penser 07 penser 08(ne) penser 09 penser NEW penser 10 penser V_7_96 penser V_7_97 permettre 01 permettre 02 permettre 03 permettre 04(s) permettre 05(s) permettre 06(s) préciser 02 préciser 04(s) préciser V_5_104 préciser V_10_161 prévoir 01 prévoir 02 prévoir 03 prévoir 04 rappeler 07 rappeler 08 rappeler 12(s) rappeler V_36R_76 reconnaître 06 reconnaître V_32R3_751 reconnaître V_36DT_213 refuser 01 refuser 04 refuser 05 refuser 08 refuser 09 refuser 11(s) refuser V_7_112 regarder 04 regarder 07 regarder V_4_581 regarder V_6_399 regarder V_7_114 regarder V_32NM_103 réussir 02 réussir 04 réussir 05 réussir 06 réussir V_5_130 réussir V_7_128 révéler 02 révéler 03 révéler 06(s) révéler 07(s) révéler V_32NM_112 savoir 02 savoir 08 savoir 09 savoir 11 souligner 04 souligner V_10_205 soupçonner 01 soupçonner 03 tenter 01 tenter 03 tenter 04(ê) tenter V_32H_685 voir 03 voir 05 voir 11 voir 12 voir 14 voir 20(s) vouloir 01 vouloir 02 vouloir 03 vouloir 05 vouloir 06 vouloir 09 vouloir 10(en) vouloir 11(s'en)

our data: annotation

PFVanim

IMP

value 0

1

Vinanim 2

3

4

5

col description sample sign.

5 max inference 1|1, 1| − 1 177

4 max under 1 polarity 1|N, 0.9| − 1 77 1-way implicatives

3 strong, not max, under 2 polarities 0.9| − 0.9 9 2-way quasi implicatives

2 strong, not max, under 1 polarity 0.9|N, N| − 0.9 8 1-way quasi implicatives

1 0 neutral, no inference not applicable or not grammatical N|N NA or UNGR

I

I

I


#sign.

78 161

signatures for either PFV+anim or PFV-anim context PFV+anim > PFV-anim ; ESPs often [+anim] only PFV darker than IMP ; stronger inferences with PFV Inferences of French ESPs

(counter-)factives, 2-way implicatives

inferential classes

neutral

19 / 38

findings

results: overview

Counts-based evidence for co-variation of inferential profile with: Semantics: outer aspect and animacy Syntax: types of embedded clauses



21 / 38

results: overview




22 / 38

outer aspect: implicatives (but not factives) have an aspect-dependent profile factives: same entailment under both polarities, signature 1|1 or −1|−1 implicatives: entailment at least under one polarity eg. 1|−1, 1|N, −0.7|−1

Factives under PFV Implic. under PFV

with IMP signature 54 77

IMP sign6=PFV sign 2 (4%) 36 (48%)

factive verbs: no influence of outer aspect implicative verbs: more change with outer aspect Ingrid Falk and Fabienne Martin


23 / 38

outer aspect: imperfective aspect weakens implicativity Observation for implicatives: PFV entailment IMP at most a defeasible inference I

I

A ce moment-là, elle a réussi à s’enfuirEMB . #Mais finalement, elle ne s’est pas enfuie. ‘At that moment, she managed to escape. But at the end, she didn’t escape.’ A ce moment-là, elle réussissait (encore) à s’enfuirEMB . OK Mais finalement, elle ne s’est pas enfuie. ‘At that moment, she ‘was still managing’ to escape. But at the end, she didn’t escape.’



24 / 38

outer aspect: imperfective aspect weakens implicativity Question about implicatives: I

Does réussir instantiate a more general pattern?

I

Is the inference generally stronger with PFV than with IMP?

77 implicative readings with PFV & IMP: IMP;weaker infer 44.2% (34)

IMP;stronger infer 2.6% (2)

no change 53.2% (41)

I

inferential profile often varies with outer aspect;

I

inference with IMP almost always weaker.



25 / 38

animacy: stronger inference with −animate subject

Reminder: I

Pierre/cela a obligé Marie à partir=obliger 02 ‘Peter/this oblige-PFV-3SG Mary to go.’

I

obliger 02 PFV/+ANIM:

0.9|N

I

obliger 02 PFV/-ANIM:

1|-0.9

Most implicatives (8 of 13) with PFV+anim and PFV-anim → PFV-anim triggers stronger inference

Most factives (33 of 42) require an animate subject.



26 / 38

results: overview




27 / 38

syntax disambiguates verbs with implicative and factive readings

Previous observations for English: I

[White, 2014] — verbs which can be either implicative or factive are disambiguated by the type of clauses they embed: I I

I

remember that (factive) remember to (implicative)

[Landau, 2001] — implicatives I I

do not take finite (that)-clauses take infinitival complement clauses



28 / 38

syntax disambiguates verbs with implicative and factive readings French: 20 verbs w. (41) factive and (45) implicative readings ; clear differences in the embedded clauses accepted by implicative vs. factive readings: 8 7 6 5 Factive Implicative

4 3 2 1 0 +QUE


+INF

+INF-QUE


-INF+QUE

29 / 38

conclusion and outlook

this study I

Set up of a (small) seed lexicon for French I I

based on inferential properties of ESPs capturing the effect of ESPs → certainty and polarity of the embedded event

I

Can be used to automatically determine factuality of embedded events

I

Support of hypotheses for French about ESP inferences: 1. implicatives, but not factives have an aspect dependent inferential profile in French 2. − implicativity with IMP aspect 3. − implicativity with + animate subject 4. syntactic type of embedded clauses: implicatives : + infinitive clause, − tensed clause factives : − infinitive clause, + tensed clause



31 / 38

to do

I

Use annotated data as seed I

I

to identify semantic/syntactic properties characteristic of inferential classes to look for similar candidates

I

Check hypotheses on larger data sets

I

(Semi-)automatically determine inferential signature?



32 / 38

references

references I

[Bittar, 2010] Bittar, A. (2010). Building a TimeBank for French: a reference corpus annotated according to the ISO-TimeML standard. PhD thesis, Paris 7. [Bittar et al., 2011] Bittar, A., Amsili, P., Denis, P., and Danlos, L. (2011). French TimeBank: an ISO-TimeML annotated reference corpus. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies: short papers-Volume 2, pages 130–134. Association for Computational Linguistics.



34 / 38

references II [Constant and Tolone, 2010] Constant, M. and Tolone, E. (2010). A generic tool to generate a lexicon for NLP from Lexicon-Grammar tables. In Gioia, M. D., editor, Proceedings of the 27th international congress on lexicon and grammar (L’Aquila, 10-13 september 2008), volume 1 of Lingue d’Europa e del Mediterraneo, Grammatica comparata, pages 79–93. Aracne. [François et al., 2007] François, J., Le Pesant, D., and Leeman, D. (2007). Présentation de la classification des Verbes Français de Jean Dubois et Françoise Dubois-Charlier. Langue française, 153(1):3–19. [Hacquard, 2006] Hacquard, V. (2006). Aspects of modality. PhD thesis, Massachusetts Institute of Technology. Ingrid Falk and Fabienne Martin


35 / 38

references III [Karttunen, 1971] Karttunen, L. (1971). Implicative Verbs. Language, 47(2):340–358. [Karttunen, 2016] Karttunen, L. (2016). Presupposition: What went wrong? Semantics and Linguistic Theory, 26(0):705–731. [Karttunen et al., 2016] Karttunen, L., Cases, I., and Supaniratisai, G. (2016). A Learning Corpus for Implicatives. Presentation at the Semantics and Pragmatics Group meeting. [Landau, 2001] Landau, I. (2001). Elements of control: Structure and meaning in infinitival constructions, volume 51. Springer Science & Business Media. Ingrid Falk and Fabienne Martin


36 / 38

references IV [Martin and Schäfer, 2012] Martin, F. and Schäfer, F. (2012). The modality of offer and other defeasible causative verbs. In Proceedings of the 30th West Coast Conference on Formal Linguistics, Somerville: Cascadilla Proceedings Project, pages 248–258. [Nairn et al., 2006] Nairn, R., Condoravdi, C., and Karttunen, L. (2006). Computing relative polarity for textual inference. In Proceedings of the Fifth International workshop on Inference in Computational Semantics (ICoS-5), pages 20–21. [Pustejovsky et al., 2005] Pustejovsky, J., Knippen, R., Littman, J., and Saurí, R. (2005). Temporal and event information in natural language text. Language resources and evaluation, 39(2):123–164.



37 / 38

references V

[Saurí and Pustejovsky, 2009] Saurí, R. and Pustejovsky, J. (2009). FactBank: a corpus annotated with event factuality. Language resources and evaluation, 43(3):227. [Saurí and Pustejovsky, 2012] Saurí, R. and Pustejovsky, J. (2012). Are you sure that this happened? assessing the factuality degree of events in text. Computational Linguistics, 38(2):261–299. [White, 2014] White, A. S. (2014). Factive-implicatives and modalized complements. In Proceedings of the 44th annual meeting of the North East Linguistic Society, pages 267–278.



38 / 38