Annotation guidelines
corpora annotated for multiword expressions

---

Multi-verb constructions (MVC)

Multi-verb constructions (MVC) constitute a quasi-universal category. They are VMWEs composed by a sequence of two adjacent verbs (in a language-dependent order), a functionally governing verb V-gov (also called a vector verb) and a functionally dependent verb V_dep (also called a pole/polar verb), which have the following characteristics:

1. They usually have the same subject.
2. They usually denote actions that are closely connected and may be seen as part of the same event.
3. They function together as a single predicate.
4. They are unaccompanied by any explicit coordination, subordination, or dependency marker.
5. They only have a single tense, aspect and polarity value.
6. They may be idiomatic or indicate successions of events.
7. The V-gov (vector) verb is semantically delexicalized and the V-dep (polar) verb contains the core meaning of the whole. Note that V-dep might be seen as the head and V-gov as the dependent, in dependency frameworks such as Universal Dependencies, where the principle of the primacy of content words is applied.

The behavior of MVCs is very heterogeneous across languages. Therefore, most tests for the detection of MVCs are language specific. The current tests were designed for Indonesian, Hindi, Japanese and Chinese. The generalization of these tests cross-lingually is planned as future work.

MVC-specific decision tree for Hindi

• Apply Test MVC.1.BASE - [MVC-STRUCT-BASE: V-dep is non finite and V-gov bears inflection?]
• It is not a VMWE, exit
• Apply Test MVC.3.KAR - [INS-REDIRECT-KAR: kar or ke appears just after V-dep?]
• Apply Test MVC.6 - [MANNER: V-gov indicates the manner/means/direction of V-dep?]
• It is a manner serial verb, not a VMWE, exit
• Apply Test MVC.7 - [REASON: V-gov indicates the reason for V-dep?]
• It is a reason serial verb, not a VMWE, exit
• Apply Test MVC.8 - [SEQ: V-gov and V-dep bound by temporal sequence?]
• It is a temporal sequence serial verb, not a VMWE, exit
• Apply Test MVC.9 - [SIMULT: V-gov+V-dep express rapid and simultaneous actions?]
• It is a serial verb expressing simultaneous actions, not a VMWE, exit
• Continue to the next test
• Apply Test MVC.10 - [LIGHT: V-gov in the closed list of light verbs?]
  • Annotate as MVC
  • Apply Test MVC.13 - [V-LEX: V-dep refers to the same event/state as V-gov+V-dep?]
  • It is not a VMWE, exit
  • Annotate as an MVC

MVC-specific decision tree for Chinese

• Apply Test MVC.2.ASPECT - [INS-DISCARD-ASP: V-gov can take un aspect marker –le or –guo?]
• It is not a MVC, exit
• Apply Test MVC.5 - [MODAL: V-gov is a modal or an auxiliary verb?]
• It is not a MVC, exit
• Apply Test MVC.6 - [MANNER: V-gov indicates the manner/means/direction of V-dep (or vise versa)?]
  • It is not a MVC, exit
  • Apply Test MVC.7 - [REASON: V-gov indicates the reason for V-dep (or vise versa)?]
  • It is not a MVC, exit
  • Apply Test MVC.9 - [SIMULT: V-gov+V-dep express rapid and simultaneous actions?]
  • It is not a MVC, exit
  • Apply Test MVC.4 - [SHARE-ARGS: V-gov and V-dep share arguments?]
  • Annotate as an MVC
  • It is not a MVC, exit

MVC-specific decision tree for Indonesian and Japanese

TODO (in the meantime, follow the tests one by one)

MVC-specific decision tree for any other language

• Apply directly Test MVC.13 - [COMP: V-dep refers to the same event/state as V-gov+V-dep?]
• It is not a VMWE, exit
• Annotate as an MVC

Test MVC.1 - [MVC-STRUCT] MVC-like structure

Does the candidate respect the necessary structural (language-dependent) requirements for an MVC?

Hindi

Test MVC.1.BASE [MVC-STRUCT-BASE]: Is V-dep non finite and does V-gov carry the tense, aspect and agreement inflections?

• continue to the next test
n.a.
n.a.
n.a.
• it is not an MVC
n.a.
n.a.
n.a.

Japanese

Test MVC.1.IMORPH: Does the first verb (V-dep) contain the i-morph suffix?

• continue to the next test
n.a.
n.a.
焼きついyakitsuisear into one's mind → the first verb 焼き yakiburnis inflected in the i-morph ending
n.a.
n.a.
• it is not a MVC
n.a.
n.a.
n.a.

Any other language

Go to the next test

Test MVC.2 - [INS-DISCARD] Insertion which discards

Does the candidate sequence appear, or could it appear, with an affix, particle or another external (non-lexicalized) material (depending on the language) which indicates that this candidate is a regular combination and should be discarded?

Chinese

Test MVC.2.ASPECT - [INS-DISCARD-ASP]: Can the aspect marker 了 -leperfective or 过 -guoprovide the meaning of the prefix be inserted between between V-gov and V-dep (or the opposite)?

• it is NOT an MVC
  我wǒ I看出来 kànchūlái figure out→ 我看wǒkàn I see了le aspect marker出来 chūlái exit→ The insertion of the aspect marker 了 le aspect markeris grammatically sound
• continue to next test
  我wǒ I听说tīngshuō heard → *我听wǒtīng I heard了 le aspect marker说 shuō say→ The insertion of the aspect marker 了 le aspect marker leads to ungrammaticality in the phrase

Indonesian

Test MVC.2.PRON - [INS-DISCARD-PRON]: Can a pronoun like dia he/she be inserted between the first [AS: between V-gov and V-dep or the opposite?] and second verb?

• it is NOT an MVC
n.a.
n.a.
n.a.
• continue to next test
n.a.
n.a.
n.a.

Test MVC.2.CLAUSE - - [INS-DISCARD-CLAUSE]: Can a that-clause like bahwa that, or a whether-clause like apakah whether be inserted between the first and second verb [AS: between V-gov and V-dep or the opposite?], where the first verb [AS: V-gov?] is a saying verb like mengatakan say or an asking verb like menanyakan ask?

• it is NOT an MVC
n.a.
n.a.
n.a.
• continue to next test
n.a.
n.a.
n.a.

Test MVC.2.PURPOSE - [INS-DISCARD-PURP]: Can untuk for/to be inserted between the first and second verb [AS: between V-gov and V-dep or the opposite?]?

• it is a purpose serial verb, not an MVC
  n.a.
  Saya Ibersiap pergi get ready to go= SayaI bersiap untuk pergi get ready for the purpose of going→ The insertion of untuk for/to is grammatically sound and does not change the meaning of the sentence. Although it is possible to insert untukfor/to between first and second verb, it is usually unnecessary and omitted.
  n.a.
  n.a.
• continue to next test

Japanese

Test MVC.2.HONOR - [INS-DISCARD-HONOR]: Is the first verb [AS: V-gov or V-dep?] preceded by the honorific particle お o and is the second verb する/できるsuru/dekiru?

• it is NOT an MVC, but an honorific construction.
  n.a.
  n.a.
  お-話し-する o-hanasi-suru I humbly talk
  n.a.
  n.a.
• continue to next test

Any other language

Go to the next test

Test MVC.3 - [INS-REDIRECT] Insertion which redirects

Does the candidate sequence appear with an affix, particle or another external (non-lexicalized) material (depending on the language) which indicates that a particular test should be applied next?

Hindi

Test MVC.3.KAR - [INS-REDIRECT-KAR]: Does conjunctive participle kar or ke appear attached to or immediately after V-dep?

• Go directly to test MVC.6 [MANNER].
n.a.
n.a.
n.a.
• Go directly to test MVC.10 [LIGHT].
n.a.
n.a.
n.a.

Any other language

Go to the next test

Test MVC.4 - [SHARE-ARGS] Shared arguments

Do V-gov and V-dep share arguments?

• it is an MVC
n.a.
• it is not an MVC
  n.a.

Test MVC.5 - [MODAL] Modal or auxiliary verb

Chinese

Is V-gov a modal or an auxiliary verb?

• it is NOT an MVC
  n.a.
  n.a.
  可以 kéyǐcan, 可能 kěnéngmight, 会 huìwill, 必须 bìxūmust, 需要 xūyàoneed to, 要 yàowant to, 能 néngable to, 应该 yīng gāishould
• continue to next test
n.a.
n.a.

Any other language

Go to the next test

Test MVC.6 - [MANNER] Manner verb

Chinese, Hindi, Indonesian, Japanese

Does V-gov indicate the manner or means (and possibly a direction) of the action expressed by V-dev (in Chinese: or vice versa)?

• it is a manner serial verb, not an MVC
  n.a.
  n.a.
  us-ne ciikh-kar mujh-e bulaa-yaa He-erg yell-ConjPpl I-dative call-perf he called me by screaming
  pulang melalui return-home pass-through go home by passing through (a place)
  投げ込み nage komi throw go in throw into
  なぐり殺し naguri korosi punch kill kill by punching
  走进来 zǒu jìnláiwalk enter walk into (a place)
• continue to next test

Any other language

Go to the next test

Test MVC.7 - [REASON] Reason verb

Hindi and Chinese

Does V-gov indicate the reason of the action expressed by V-dep (in Chinese: or vice versa)?

• it is a reason serial verb, not an MVC
  n.a.
  n.a.
  vo melaa jaa-kar khush hu-aa he fair go-ConjPpl happy become-perf he got happy having gone to the fair
  n.a.
• continue to next test

Any other language

Go to the next test

Test MVC.8 - [SEQ] Temporal sequence

Hindi, Indonesian, Japanese

Are the verbs bound by a temporal sequence?

• it is a sequential serial verb, not an MVC
  n.a.
  n.a.
  us-ne gilaas banaa-kar bec-aa he-erg glass make-ConjPpl sell-perf having made the glass, he sold it
  bersiap pergi prepare go prepare in order to go (somewhere) → the first verb must happen before the second verb happens, otherwise the sentence will not make sense.
  夫人が最初にfujin ga saisho ni the wife first叩き起こさtataki okosa hit to awakenれre verb suffix != #夫人が最初にfujin ga saisho ni the wife first起き叩さtataki okosa hit to awakenれ re verb suffix→ The two verbs 叩き tataki hitand 起こさ okosa awakenare bound by temporal sequence, such that if the order is switched, the sentence does not make sense.
  n.a.
• continue to next test

Any other language

Go to the next test

Test MVC.9 - [SIMULT] Simultaneous actions

Do the verbs indicate rapid and simultaneous actions (without resorting to a coordination conjunction)?

• it is a serial verb expressing simultaneous actions, not an MVC
  n.a.
  n.a.
  berlari menuju run head-towards run and go towards
  n.a.
• continue to next test

Test MVC.10 - [LIGHT] Light verb

Hindi

Does V-gov belong to a closed list of light verbs: aa come, baiTh sit, chal go, chuk finish, choR leave, Daal throw, de give, ja go, jataa declare, khaa eat, lagaa put, le take, maar hit, paa get/obtain, paRh fall, rakh keep, uTh rise?

• it is a (light) MVC
  n.a.
  n.a.
• continue to next test

Any other language

Go to the next test

Test MVC.11 - [PREP-LIKE] Preposition-like verb

Chinese

[Hongzhi Xu: this test is not very clear and is only specific to one particular MVC (it should probably be deleted in future editions)] Is the second verb in the candidate [AS: V-gov or V-dep?] a preposition-like verb like 成 chéng become?

• it is a preposition-like MVC
  n.a.
  n.a.
  n.a.
  排列成 páiliè chéng arrange become arrange into (something)
• continue to next test

Any other language

Go to the next test

Test MVC.12 - [NOUN-LIKE] Noun-like verb

Japanese

Are any of the components [AS: V-gov or V-dep?] in the candidate noun-like arguments?

• it is a deverbalized V1/V2 MVC
  n.a.
  (JA) 響き渡る hibiki wataru echo spread-widely reverberate → The first verb is a noun-like argument of the second verb [deverbalized V2]
  聞き違え kiki chigae listen be-different mishear/misunderstand → The second verb is a noun-like argument of the first verb [deverbalized V1]
  n.a.
  n.a.
• continue to next test

Any other language

Go to the next test

Test MVC.13 - [V-LEX] Lexical inflexibility

Does a regular replacement of V-dep by a related verb taken from a relatively large semantic class lead to ungrammaticality or to an unexpected change in meaning?

• it is an MVC
it will make do → #it will make build/solve/construct
hello quiere decir "hola" en inglés "hello" want say "hola" in English "hello" means "hola" in English → #"hello" quiere comunicar/gritar "hola" en inglés "hello" want communicate/shout "hola" in English
j'ai laissé tomber la présentation I have let fall the presentation I gave up on the presentation → #j'ai laissé commencer/lancer/interrompre la présentation I have let start/launch/interrupt the presentation
ce mot veut dire autre chose this word wants say other thing this word means something else → #ce mot veut chuchoter/communiquer/crier autre chose this word wants whisper/communicate/scream another thing
ik heb ze leren kennen I got to know them→ #ik heb ze leren lezen/schrijven
dał jej popalić he let hr smoke he made things hard for her → #dał jej popić/podymić/pofajczyć he let her drink/smoke
n.a.
• it is not an MVC
  it will make me think → it will make me build/solve/construct
  quiero leer tu tesis want.I read your thesis I want to read your thesis → quiero adquirir/descargar/imprimir tu tesis want.I acquire/download/print your thesis I want to get/download/print your thesis
  je l'ai laissé finir la présentation I him have let finish the presentation I let him finish the presentation → je l'ai laissé commencer/lancer/interrompre la présentation I him have let start/launch/interrupr the presentation
  ce garçon veut dire autre chose this boy wants say other thing this boy wants to say something else → ce garçon veut chuchoter/communiquer/crier autre chose this boy wants whisper/communicate/scream another thing
  ik heb mijn trui laten wassen I had my sweater washed→ ik heb mijn trui laten strijken/verven/maken I had my sweater ironed/dyed/repaired
  dał jej pospać he let her sleep→ dał jej odpocząć/poleżeć he let her rest/lay
  n.a.

---