Ling 567 Lab 3
Nigel Kilmer and Woodley Packard
Northern Frisian (frr)

The test suite is 'testsuite.txt', and the parsed profile is 'tsdb/home/lab3'.

===========================
1. Documentation of Choices
===========================

Frisian has a single adverb 'ek' for clausal negation.  The negator generally appears clause-finally, e.g.:

ja   kum  miaren   ek
they come tomorrow not
they aren't coming tomorrow

verpflegungs.jil fairst du ek
caretaking.money get.2SG you not
you aren't getting any expense money

The adverb 'ek' also very commonly appears as a left-modifier of other constituents, e.g. 'ek muar' (not much/more), 'ek wech' (not away), including non-finite verbs, e.g.:

gur, sair di tiiger, ik wel di det lop ek iit
good said the tiger  i will you this time not eat
Alright, said the tiger, I won't eat you this time

In some cases it is unclear whether the negator is modifying a complement or the verb:

naan, dit es ek min aart
no,  that is not my way
no, that is not my way

For now we are supposing that the negator is always either a left-modifier or clause-final.  One potentially problematic aspect of this analysis is that there are no attested examples of 'ek' appearing to the left of the finite verb (in the forfelt).

Definiteness of NPs is marked by the definite articles 'di' (m/f) and 'dit' (n), which agree in gender.  As the object of a preposition in cases where the referent is unambiguous, the definite article is optional.  For now we have elected to mark determiners as optional on all nouns (except pronouns and certain closed-class nouns like 'fuul' = 'much').

Demonstratives in Frisian are also articles, agreeing in gender.  There is a three-way proximity distinction, with the two extremes corresponding to "this" (= "dijir" and "ditjir") and "that" (= "didiar" and "ditdiar") in English, and a third option "des"/"det" whose meaning is apparently less committal about proximity.  We have lexical entries for all six demonstrative determiner forms, corresponding to three different predicates.

All arguments in Frisian are obligatory, with the exception of second person singular pronoun subjects, which are optional:

jil fairst ek
money get.PRES-2SG not
you're not getting any money

In the questionnaire, we indicated that object dropping never occurs, but subject dropping occurs with all verbs, with the identity of the subject always marked on the verb (in the form of person/number agreement) regardless of whether the subject is dropped or not.  We created a subject dropping context indicating that the subject NP must be 2SG.

Frisian exhibits agreement of determiners with their nouns in gender and number, e.g. "di hingster" (the horses / common) but "dit heest" (the train / neuter).  This is implemented as features on the determiners.  Additionally, subjects must agree with the verb in person and number.  This agreement is implemented by features specified by lexical rules in an obligatory verb inflection position, together with features on noun/pronoun lexemes and introduced by obligatory inflection for number on non-pronoun nouns.

There is a preterite tense indicated by the affix '-t' on a verb, which we implemented with lexical rules that constrain [TENSE past].  Note that the affix '-t' is quite ambiguous in Frisian; in 3SG usage, it is impossible to distinguish preterite tense from present tense, hence we have only one 3SG lexical rule, which leaves TENSE unconstrained.  The 2SG affix '-st' is also ambiguous between present tense and preterite tense, hence we have only one 2SG lexical rule.  For 1SG and NON-SG (i.e. dual or plural), we subdivide the lexical rules into past and nonpast varieties, since the TENSE is predictable from the form.

There is apparently additionally a future tense formed with the auxiliary 'skel', which we implemented with an auxiliary verb that selects for a verbal projection with [TENSE nonpast] and [FORM nonfinite].  This tense is not commonly used, however; usage of the present tense combined with a temporal adverb has largely taken over its function.

Verbs are normally in the imperfective aspect.  The perfective aspect is formed with an auxiliary 'haa' and the past participle, which is also formed with the affix '-t' on a verb (another lexical rule).  We created an auxiliary verb 'haa' which selects for a verbal projection with [ASPECT perfective] and [FORM nonfinite].  The perfective aspect and past tense can be combined to form a pluperfect.

We also rewrote the PRED values in our lexicon to use English and added several noun and verb lexemes.

=======================================
2. Untreated Phenomena in the Testsuite
=======================================

The sentential negation examples in our resource include relative clauses and intensifiers.  We built reduced examples that should be within the grammar's coverage; we believe we have the grammaticallity judgements correct on them, but of course it's hard to know for sure.

Some of the properties of definiteness in Frisian only appear in prepositional phrases.  We do not yet have a treatment of prepositional phrases.

Some of our examples showing argument optionality are yes-no questions.  These are not yet treated by our grammar, so they do not parse:

heest gur slöpen
have-2SG good sleep-PST.PTCP
did you sleep well?

The examples of demonstrative determiners in our reference tend to feature non-referential subjects, e.g.:

diar laap di hingster en di fölken ön di guart
diar laap di hingst-en en di fölk-en ön di guart
there run the horse-PL and the foal-PL in the yard
there are horses and foals running around in the yard

==========================
3. Coverage of the Grammar
==========================

The testsuite that we used for this lab contains 111 items.

Our grammar analyzes 46 / 68 positive items in the test suite, for a coverage of 67.6%.

We parse 0 / 43 negative items, for 0% overgeneration.

Of the 22 positive items that we cannot parse, one has full lexical coverage but uses the negator 'ek' as a left modifier of a constituent rather than sentence finally.  The rest have lexical gaps.

The lexical gaps are adverbs (e.g. 'mal' = 'very'), ditransitive verbs ('dö' = 'give'), and prepositions (e.g. 'bi' = 'by'). So far we do not know how to define any of these in the customization system.

=======================================
4. Coverage of the Customization System
=======================================

It's not completely clear whether the sentential negator 'ek' should be restricted to appear clause-finally.  There are many, many examples where it appears in other places (always right of the finite verb), but it may be modifying another constituent in those cases.  There doesn't seem to be a customization option to allow it to appear in arbitrary positions to the right of the finite verb.
