Keywords

1 Introduction

Quechua is the language of the Incas, an ancient civilization in South America. It’s spoken by over 6 million people in Peru, Equator, Bolivia, and Argentina.

Concerning the verbs we notice two important characteristics: All the Quechua verbs are regular verbs, typologically speaking it is a SOV language.

We show in this paper how to generate new Quechua verbs through the derivation of simple verbs. We show also how to annotate them using certain morphological grammars programmed with the help of NooJ (Silberztein 2003, 2010). To reach this objective we should go through the following steps: first, we identify the basic canonical conjugated verbal form and isolate two classes of suffixes, the Inter positional suffixes IPS (Duran 2009, 2013), which will allow us to obtain new verbs and the post-posed suffixes PPSFootnote 1 which will present different modalities of a given verb without changing the part of speech. Parker calls it “modal suffix system” (Parker 1969). Then we describe in detail some of these generating paradigms in the NooJ environment (Silberztein 2016). We show some details of the programs that we have built to annotate the inflected verbs, then we describe how we can use this method to obtain the Quechua-French translation of these verbs with the help of automatically generated annotations.

2 The Basic Canonical Conjugated Form

The main canonical conjugation form for any transitive verb follows the scheme

figure a

In which the person is marked but the time is not as is shown in the following table (Table 1):

Table 1. The unmarked-time basic conjugation of a verb
Full size table

We will use the symbol PR for the set of these basic endings:

PR = (NI, NKI, N, NCHIK, NIKU, NKICHIK, NKU).

The general, non-future, verbal canonical form is:

figure b

Where IPS symbolizes any interposed suffix and PPS symbolizes any postposed suffix: e.g.:

llamka-ri-ni

I begin to work

(1 IPS before the NI ending)

llamka-ni-raq

I first worked

(1 PPS after the NI ending)

A typical verbal form contains one to two agglutinated IPS preceding the ending and one to two PPS following the ending like in the following examples:

llamka-chi-ni-raq

I first made him work

(1IPS before, and 1 PPS after the NI ending)

llamka-na-nchik-raq-mi

We have to do the work putting everything aside

(1IPS before the NCHIK ending and 2 PPS after the NCHIK ending)

In the last two inflected forms ni and NCHIK act as fixed points.

As a matter of fact, a remarkable property during the inflection of any transitive verb is that the PR endings behave as fixed points around which IPS or PPS suffixes may be agglutinated to obtain a verbal form. The following table shows some more examples for the ni ending:

The PR ending acts as a fixed point.

For the other endings, we may obtain their corresponding form replacing the NI ending by the one we want to use, leaving unchanged the rest of the form.

Thus if we represent by Fni the NI form: stem + IPS + ni + PPS.

And by Tnki the transformation that replaces the NI ending by the NKI ending. We will have:

figure c

Where IPSFootnote 2 may be an agglutination of 1 to 4 interposition suffixes and PPSFootnote 3 may be an agglutination of 1 to 3 postposition suffixes.

3 Verbal Morphology

3.1 PPS Mono Suffixation

The mono suffixed inflection of a transitive verb including the future conjugation using 1 PPS or 1 FSFootnote 4 (future conjugation endings) are obtained applying the following NooJ grammar:

figure d

Where

SPP1_V is the paradigm for the endings ending in a vowel and SPP1_C for those ending in a consonant

figure e

Giving a total of 210 inflected forms containing one postposition suffix as shown in the following Fig. 2.

The following is a sample of these inflected forms:

figure f

3.2 IPS Mono Suffixation

On the other hand, the mono suffixation of verb stems by IPS suffixes are very productive compared to the PPS derivations. They give rise to new verbs as in these examples:

asiy

«laugh»

> asi-ri-y

to smile

ripuy

«go away»

> ripu-ku-y

to move

rakiy

«to split»

> raki-naya-y

to feel like splitting

The resulting forms can be conjugated as if it were a simple verb. Figure 3 shows the NooJ grammar that generates them.

Below, we present an extract

  • rimaruy, rimay, V + FR = “parler” + FLX = V_SIP1_INF + PRES + INF

  • rimarquy, rimay, V + FR = “parler” + FLX = V_SIP1_INF + PAPT + INF

  • rimariy, rimay, V + FR = “parler” + FLX = V_SIP1_INF + DYN + INF

  • rimarayay, rimay, V + FR = “parler” + FLX = V_SIP1_INF + DUR + INF

  • rimapuy, rimay, V + FR = “parler” + FLX = V_SIP1_INF + APT + INF

  • rimapayay, rimay, V + FR = “parler” + FLX = V_SIP1_INF + FREQ + INF

  • rimanayay, rimay, V + FR = “parler” + FLX = V_SIP1_INF + ENV + INF

  • rimakuy, rimay, V + FR = “parler” + FLX = V_SIP1_INF + AUBE + INF

  • rimakapuy, rimay, V + FR = “parler” + FLX = V_SIP1_INF + RAS + INF

  • rimakachay, rimay, V + FR = “parler” + FLX = V_SIP1_INF + ARO + INF

  • rimaysiy, rimay, V + FR = “parler” + FLX = V_SIP1_INF + COLL + INF

The Quechua morphology allows certain agglutinations of two V_SIP2_INF and three V_SIP3_INF interposition suffixes. These new forms are also conjugable ones. They can be obtained by applying the following paradigms programmed in NooJ.

figure g

It contains more than 240 valid agglutinations.

And for three IPS agglutinated suffixes

figure h

Which contains more than 2470 valid agglutinations.

We present below some examples of ternary and more IPS suffix layer agglutinations:

figure i

3.3 The Mixed Inflections IPS + PPS

As we have seen in Fig. 1, an important property of Quechua grammar is that we can mix these two types of suffixation to generate current verbal forms. We have been able to program, with NooJ, the following paradigms of mixed agglutination:

figure j
Fig. 1.
figure 1

The PR ending acts as a fixed point

Full size image
Fig. 2.
figure 2

Inflected forms bearing one interposed suffix

Full size image
Fig. 3.
figure 3

The NooJ grammar that generates derived verbs using one suffix IPS

Full size image

e.g.: miku-chi-ni-raq

figure k

e.g.: miku-chka-ni-raq-mi

figure l

e.g.: miku-chi-chka-ni-raq-mi

Where the symbols stand for V_MIX12: mixed verbal agglutination of 1 IPS and 2 SPP.

SIP1_PR_V: Agglutination with 1 IPS and conjugated following the PR scheme, etc.

We obtain, for the case of the verb rimay/to talk, 289 413 mixed and conjugated forms as shown in Fig. 4.

Fig. 4.
figure 4

Total of mixed and conjugated forms for the verb rimay

Full size image

3.4 Verbalizers

An interesting strategy of Quechua is to generate verbs using a set of verbalizer suffixes. They may be added to nouns as well as to adjectives; these are:

  • The infinitive –y,

  • The transformative (creation or destruction) transitive -chay,

  • The transformative (to become) intransitive –yay.

Each one can be suffixed to certain classes of nouns or adjectives. Some nouns can be verbalized by only one of these suffixes and some nouns accept verbalization with two or all three of them: e.g.:

figure m

The following NooJ grammars may be utilized to generate verbs:

  • YAY = yay/V; # derives some type of nouns and adjectives into verbs chukru =>chukruyay to toughen

  • CHAY = chay/V; # derives some type of nouns and adjectives into verbs puka => pukayay/to become red

  • Y = y/V;

3.5 The Quechua-French Translation Problem

One of the objectives of our long-term project, of machine translation from French to Quechua, is to build bilingual French-Quechua dictionaries of each PoS (Nouns, Verbs, Adjectives, etc.).

Concerning the Quechua lexicon of simple verbs, it contains around 1500 entries. It is not enough to get the translation of the 25 000 French verbal senses contained in Dubois & Dubois-Charlier dictionaryFootnote 5. For instance, the verb “tourner”/to turn, has 27 meanings in LVF (Dubois and Dubois-Charlier 2007; François et al. 2007); how to find its translation knowing that the only possible equivalent simple verb in Quechua is muyuy.

4 Generation of ALUc

But, as we have just seen above, Quechua has a very productive strategy: Derivation V > V, by suffixation. We have programmed the following paradigms in NooJ and generated 43 394 new Quechua verbal senses (Fig. 5 . Total of mixed and conjugated forms for the verb rimay), ALUc, that can be conjugated as a simple verb.

Fig. 5.
figure 5

Total of mixed and conjugated forms for the verb rimay

Full size image

Total generated new verbal senses

An extract of which appear in the Fig. 6.

Fig. 6.
figure 6

Extract of the generated new verbal senses

Full size image

Do all of these ALUc have an actual meaning? In the affirmative case, the expected translation FR > QU of verbs of the Dubois & Dubois-Charlier dictionary seems to be feasible.

How to figure out the meaning of those derived forms which are not so familiar in everyday speaking or in the written corpus? When we apply the suffix –chi- to the verb muyuy/to turn we obtain muyuchiy/make it rotate, but is it possible to translate muyu-chi-tamu-y or muyu-kapu-lla-y avoiding long periphrases?

In order to aboard such a challenge, we have done the following steps: First we have identified and formalized the agglutination of IPS suffixes in the form of NooJ morphological grammars using one, two and three suffixes and applying some matrix methods to calculate the combinatorial. Second, we have inventoried the semantic characteristics of each IPS suffix and construct an indexed table. Third, we have constructed several NooJ grammars, like the ones appearing below, showing these modalities as annotations:

figure n

4.1 Modalities of Enunciation of IPS Suffixes

We have made an inventory of all the main modal and aspectual meanings of the 27 inter positional derivational suffixes SIP. We show in Fig. 7 an extract of these descriptions.

Fig. 7.
figure 7

Modalities of enunciation of IPS suffixes

Full size image

Parsing these grammars gives us several propositions as possible translations of each derived ALUc (Fig. 8):

Fig. 8.
figure 8

A glossed approach for the translation of derived verbs

Full size image

For example: among the list of derivations of the verb rimay/to talk, it appears rimaykuy. It is the output of the inflection obtained when we apply to it the suffix –yku-. Its corresponding NooJ grammar has generated the following annotation in French:

figure o

We may use these automatic annotations in order to figure out the translation that feats better the ALUc.

We choose (manually): (the subject) talks to someone courteously, carefully, friendly ,

Thus we’ll have “to greet” as the nearest translation of rimaykuy.

Let’s see an example in the sense French > Quechua. For one of the verbal senses of the verb “tourner”/to turn in French. We choose one of the 27 meanings of “tourner” appearing in the Dubois & Dubois-Charlier dictionary.

Meaning 13: tourner «Il indique que l’action se réalise de façon persistante»

To translate this verb, into Quechua, we search among the corresponding annotations in the output of inflections obtained when we apply the suffix –paya- to the verb muyuy, (that the modalities of paya appear in Fig. 7). Its corresponding NooJ grammar has generated, among 27 others, the following annotation in French:

figure p

We choose manually:  (the subject) “_tourne sans répit, répetition fréquente” ,

Which approaches better to meaning 13. Thus we’ll have

muyupaayay: tourner de façon persistante (rotation)

We have mentioned before that the lexicon of simple verbs of Quechua contained around 1500 entries. We have found many derived verbs as translation of simple French verbs (like sourire/asi-ri-y coming from asiy/rire “to laugh”; bougonner/rima-pa-ku-y coming from rimay/parler “to talk”, etc.). Many others come from verbalizations of nouns or adjectives like: aplanir/pampa-cha-y coming from the noun pampa/plane; appauvrir/wakcha-ya-y/“to become poor” coming from the adjectif wakcha/orphelin, pauvre “poor”. However, for the translation of several thousands of French verbs we have utilized our annotation method described above. In Fig. 9 we present an extract of our newly constructed electronic French-Quechua verb dictionary containing more than 8600 entries (Duran 2017).

Fig. 9.
figure 9

Extract from Electronic French-Quechua (See Duran (2017).) verb dictionary

Full size image

5 Conclusion and Perspectives

We have constructed several thousands of morpho syntactic paradigms for the inflection and derivation of verbs,

We have obtained the annotated verbal inflexions which serves as a database for the translation of several thousand french verbal senses into Quechua,

We plan to enhance the number of paradigms containing multi suffixed derivations to enlarge the FR-QU dictionary.