Exam 1: Some Answers

1. The regular past tense morpheme of English, spelled -ed, has three pronunciations depending on the consonant or vowel that precedes it. If it is preceded by /t/ or /d/, it is pronounced /@d/, where /@/ is an ascii representation of a central vowel (schwa). Otherwise, if it is preceded by a voiceless consonant (/p/, /f/, /s/, /ch/, /sh/, or /k/), it is pronounced /t/. Otherwise it is pronounced /d/. We can view the past tense morpheme as having the underlying (lexical) form /d/ and being subject to two rules, (1) a rule which inserts the vowel /@/ between a preceding /t/ or /d/ and the past-tense /d/ and (2) a rule which changes /d/ to /t/ following a voiceless consonant. Rule (1) must precede rule (2). Thus for ripped, we have

   rIp+d -> rIpt (by rule (2))
   

   and for fitted, we have

   fIt+d -> fIt@d (by rule (1))
   

   • Explain how these rules would be implemented in the two-level (KIMMO) framework.
   The rules require the following feasible pairs: K:, T:, 0:@, d:t, d:*, +:0, where K represents the set of voiceless consonants, T represents {t,d}, + represents a morpheme boundary, @ is the high central vowel (schwa), and * is the wild card symbol. In the rules # represents a word boundar, and *:* represents any pair other than those specified on other arcs out of a node.

   This rule inserts the vowel /@/ in the appropriate environment.

   d:t <=> K: (+:0) ___ #:#
   

   

   0:@ <=> T: (+:0) ___ d: #:#
   

   This rule changes /d/ to /t/ in the appropriate environment.

   

2. In English, indefinite articles (a(n), some) must agree with the nouns they modify. Singular count nouns such as cat and idea take a(n) while mass and plural nouns such as cheese, air, and cats take some.
   • Explain how a feature-based grammar would handle it. You needn't include all of the HPSG structure (SYNSEM, DTRS, etc.).

   Here is a set of lexical entries and rules similar to HPSG structures, but greatly simplified. MC represent the MASS-COUNT feature, which has two possible values, mass/plur and count-sing.

   NP rules
   
   [CAT np
    HEAD [CAT n MC count-sing]
    SUBCAT ([CAT det MC count-sing])]
   
   [CAT np
    HEAD [CAT n MC mass/plur]]
   
   [CAT np
    HEAD [CAT n MC mass/plur]
    SUBCAT ([CAT det MC mass/plur])]
   
   lexical entries
   
   [PHON <cat>
    CAT n
    MC count-sing]
   
   [PHON <cats>
    CAT n
    MC mass/plur]
   
   [PHON <cheese>
    CAT n
    MC mass/plur]
   
   [PHON <a>
    CAT det
    MC count-sing]
   
   [PHON <some>
    CAT det
    MC mass/plur]
   
   [PHON <the>
    CAT det]
   

4. Both morphology and syntax are concerned with structure, the structure of words in the case of morphology, the structure of sentences in the case of syntax. As we have seen, quite different mechanisms are called for in the two cases. Contrast the task of processing morphology with that of processing syntax, and discuss why different mechanisms are required. Give examples.
(a sketch of an answer; weak on examples)

Both morphology and syntax deal with the relationship between a "deep" and a "surface" structure. Parsing (analysis) is the process of going from surface to deep, and generation is the process of going from deep to surface in each case. In both cases it is possible to think of the deep level as consisting of elements (morphemes or phrases) which are unordered and the surface level as consisting of elements which have their order specified. Both morphology and syntax may be treated together with or separately from semantics since the basic elements are meaningful. In both cases, there is no clear isomorphism between semantic structure and the structure that emerges on the surface. For both syntax and morphology, the form that the elements take depends to some extent on the environment in which they occur. For syntax, this is especially the case for agreement phenomena. For morphology, it is a matter of the phonology of the morphemes which come together. In both syntax and morphology, a single element, the head, dominates in a structure.

At the same time, there are differences between morphology and syntax which have implications for processing. Syntax is recursive; this means that finite-state approaches are inadequate. Morphology is, with possible minor exceptions, not recursive, so finite-state approaches are adequate. In syntax there are long-distance dependencies: the form of a word may depend on another word which is arbitrarily far away from it. This calls for techniques which go beyond simple context-free grammars, either transformations or feature-based grammars. In morphology, by contrast, the form of the elements depends only on the local phonological context.

[L645 Home]