Hup is an Nadahup language spoken in the Amazon rainforest by about 1500 people. The interesting challenges
for CCG are the language's agglutinative verb, which extends to modals and stems, as well as its lack of many
characteristic constructions seen in Indo-European languages. (This is, in fact, characteristic of American Indian languages.)
The implementation is based on a 2005 dissertation from Patience L. Epps.
Download the Hup grammar specification.
This file covers the following:
Note that the degree of coverage varies from extremely superficial to mostly superficial.
To avoid dealing with possible Unicode issues at this point, the .ccg file linked above
contains some hack-ish case alternations. Most notably, /T/ is used for the glottal stop,
and /q/ is used after consontants to signify glottalization. Vowel alternations are manifested
in a capital V. Lastly, non-Roman place alternations or accenting for vowels resulted
in somewhat ad hoc case alternations for vowels.
In Hup, these are fairly simple constructions. What should be noted is the treatment of verbal affix morphemes:
'I go' 1pl GO - decl Tin ham - Vh np s\np s\s
This parses in the quite obvious manner seen below:
Parse: s ------------------------------ (lex) Tin :- np (lex) ham :- s\np (<) Tin ham :- s
I will not show parses for all examples; all of these can be seen in the test bed.
This is a simple type change from s\np to (s\np)\np. The object argument is marked
for objecthood with an /-an/ suffix, which is treated both as a morpheme for irregular
nouns, such as pronouns, as well as
These are intransitive nouns that require an extra non-oblique case-marked argument:
'My mother died (which affected me adversely)' 1sg = mother 1sg.obj die - tel - infr - decl Tah Tin Tan naT yiT ni Vh np/np np np s\np s\s (s\np)\s s\s
Ditransitives are also affixally or verbally induced in Hup:
'I will look for lice for you' louse 2sg - obj 1sg see - coop haem Tam an Tah key nin np np np\np np (s\np)\np s\s 'I am showing the child to Mih' 1sg 3sg = child - obj mih - obj show - dynm Tah tih doT an mih an be Vy np np/np [ np ] [ np ] ((s\np)\np)\np s\s
Bracketed categories are those treated in the .ccg file with expansions.
Like many American Indian languages, Hup does not have traditionally
recognized 'control' constructions. They use different strategies to
convey personal desire, as seen in the following:
'I want you to eat' 1sg 2sg - obj eat - frust - dynm Tah ti - an waed yaeh Vy np np np\np s\np ((s\np)\np)\(s\np) s\s
Relativization is achieved using two different morphemes, both
found at the right edge of the verb. One is a dependent morpheme,
while the other is derived from the object marking morpheme seen in
the examples above. Object relativization is seen here:
'That frog they were looking at' 3pl see - perf - dep frog hid key Te p hohoh np (s\np)\np s\s (np/np)\(s\np) np
This parses in the following manner:
Parse 1: np ------------------------------ (lex) hid :- np (>T) hid :- s/@i(s\@inp) (lex) key :- s\np\np (lex) Te :- s\s (<B) key Te :- s\np\np (>Bx) hid key Te :- s\!np (lex) p :- np/np\(s\np) (<) hid key Te p :- np/np (lex) hohoh :- np (>) hid key Te p hohoh :- np
Subject relativization is seen here:
'she who had been the younger sister of snake's wife' that snake - child.mother poss sister be - perf - dep yup mqaeh taehTin nIh bahTay ni Te p
Object relativiation can also be accomplished using a second affix:
'They were searching for their toad, the one they had been looking at' 3pl.poss toad 3pl see - perf - flr - obj.rel 3pl search - dep hidnih hohoh hid key Te w an hid yohoy Vp np/np np np (s\np)\np s\s s\s (np\np)\(s\np) np (s\np)\np s\s
There is no lexical conjunct in Hup. Coordination is accomplished for nominal forms
intonationally. This means, therefore, the inclusion of some intonation in the implementation.
'they took all kinds: gourds, macucu fruit, and those tucuma fruits' all gourd=fruit %LL macucu=fruit %LL this tacuma=fruit 3pl take - dist - decl nihuT bqoTtat LL hahawtat LL nup gqobtat hid doT pid Vh np/np np np\np/np np np\np/np np/np np np (s\np)\np s\s s\s
Coordination of verbal forms is accomplished via agglutination:
'they were drinking and singing' 3pl drink - sing - perf - dec hid Teg yamhidoT Te Vh np s\np (s\np) \ (s\np) s\s s\s
Because of a lack of data regarding this example, a post-stem verbal category was
posited in the implentation. This would complement the pre-stem verbal category used
for specific modals to be seen later.
family Vstempost {
entry: (s<1> [F] \ np<2> [X nobj]) \ (s<3> [E] \ np<2> [X nobj]) : F:action(* <Actor>X:animate-being CONJ E);
}
Adverbs alternate for negation:
'he runs fast' fast 3sg run - dynm pib tih toToh Vy s|s np s\np s\s 'he doesn't run fast' 3sg run - fast - neg tih toToh pib nih np s\np s/s s\s
While the code accounts for both positive forms, it does not handle the ungrammatical
forms of an incorporated affirmative adverb or an unbound negative adverb. More will
be discussed on this in the Issues/Suggestions section.
Agglutination accomplishes multiple tasks, including those seen above. This section focuses
on a few more, but is in no way completely thorough in the potential forms the verb may take.
Applicatives are licensed by the modal pre-verb 'take', as below:
'the towel hangs from the string' string - obl towel hang - dynm tit it toaya gqaT ay np (s|s)\np np s\np np 'Ped hangs the towel form the string' ped towel take - hang - dynm string - obl ped toaya dqoT gqaT Vy tit it np np ((s\np)\np)/s\np s\np s\s np (s|s)\np
Note the free ordering of obliques. This modal pre-verb is defined in the .ccg file as follows:
family Vstempre {
entry: (s<2> [F] \ np<3> [X] ) / s<1> [E] : F:action(* <Causer>X:animate-being CONJ E);
}
What should be noted here is the similarity to the Vstempost family above.
Many of the verbal affixes available carry a great deal of extra meaning. Much of this not been
accounted for semantically, but does parse syntactically. This is because many of the affixes have
some seemingly pragmatic usages. However, the ventive suffix is treated here as a post-verbal verb
rather than a suffix order to get the conjunctive meaning seen in the gloss.
'that daughter of cirino's went and came right back, apparently' that cirino daughter go - vent - perf - infr - frust - decl nqipi cididu tog nam Tay Te aid yaeh Vh np/np np/np np s\np (s\np)\(s\np) s\s s\s s\s s\s
I did not have many opportunities to use expansions. However, I did use them for regular nouns to
show morphological alternation between object-markedness and non-object-markedness:
def normal-noun(Stem, Class) {
word Stem:NP(Class) {
*: 3rd sg nobj nobj-X sg-X;
Stem . an: 3rd sg obj sg-X;
}
}
For any third-person non-pronominal noun, therefore, I could use a single word and category rather
than two for object marking /-an/. However, the family and definition for /-an/ were still necessary
for pronominals, so the point was rather moot, and missed the interesting treatment of morphology that
CCG has to offer.
The goal of this section is to discuss shortcomings of my implementation, and possible areas for
improvement on the CCG wiki.
should have a much more in-depth treatment on how to use the hybrid logic semantics in OpenCCG in the wiki. Of course,
I'm definitely not the person to do it, but anyone who could give some brief examples about maybe constraining things
semantically or featurally would be a big help. Half the time I didn't understand the syntax of the semantics, and half
the time I think I just didn't get the semantics. A hybrid logic for OpenCCG wiki would be a big help for both parts.
this when I started, and I only have an inkling of what to do now. This will probably be revisited after my Bantu verbal
extension paper in CCG.
able to use pred values. Not that this is a huge deal, but it resulted in me keeping my old entries commented out anyway so I
could check back and make sure I was using the right predicate.
Here are a few predictions not seen directly through data that my implementation predicts:
Points 1 and 3 are made in the testbed for the .ccg file.
This work is a long way from finished, but an intersting exercise in what CCG is capable of for some very
non-traditional syntactic schemes. Let's keep on treating interesting and uncommon languages that remain
problematic for more traditional theories, both to assist in preservation and documentation as well as to
show the strength of the formalism.
This can be used to decode any examples above as well as
the various features used in the .ccg file.
pl: plural decl: declarative tel: telic, contrastive emphasis infr: inferential evidential dynm: dynamic appr: apprehensive mood obj: object marker frust: frustrative (polysemous) perf: perfective dep: dependent marker flr: filler form dist: distributive neg: negative vent: ventive poss: possessive rel: relative obl: oblique