Assignment 3: Southern Brazilian Portuguese

In this assignment, the task is to create a rule system that maps from orthographical strings in Portuguese (this will be the lexical level) to strings that represent their pronunciation (this will be the surface side). A sample mapping of written "caso" to spoken "kazu" looks like this, with, by convention, the lexical string on top and the surface string on the bottom.

Standard Portuguese orthography is not always a complete guide to the pronunciation of a word (especially in the case of the letter "x" and the vowels written "o" and "e"). As usual, we will restrict and simplify the data slightly to make the solution manageable as a class exercise. Later we will redo the same example using two-level rules.

The Facts to be Modeled

• The following description is based on the rather conservative pronunciation of Portuguese in Porto Alegre, Rio Grande do Sul, Brazil. Because the orthography is even more conservative, the rules will roughly characterize the phonological changes that have occured in one dialect since the orthography fossilized.

• Your grammar should generate from lexical strings (using apply down) like the following, written in standard Brazilian Portuguese orthography. We will limit the input to lowercase words in this exercise.

  casa cimento me disse peruca simpático braço árvore

• The surface level produced by your grammar will be a kind of crude phonemic alphabet, with the following extra symbols:

  J

  palatalized "d", similar to the phoneme spelled "j" in English "judge".

  C

  palatalized "t", similar to the phoneme spelled "ch" in "church".

  $

  alveopalatal sibilant, like the phoneme spelled "sh" in English "ship".

  L

  phoneme spelled "lh" in Portuguese "filho" (or "gli" in Italian "figlio").

  N

  phoneme spelled "nh" in Portuguese "ninho" (like the French "gn" in "digne").

  R

  phoneme written "rr" inside words, single "r" at the beginning.

  Because we have limited our input words to lowercase letters, the six special characters will appear only in surface strings, never at the lexical level. The dollar sign $ character is special in regular expressions, so precede it with a percent sign (%) to literalize it or put it in double quotes.

• The mapping from orthography (lexical side) to pronunciation (surface side) includes the following:

  • The orthographical (lexical-side) "ç" is always pronounced /s/; in other words, a "ç" on top always corresponds to an "s" on the bottom. (In these explanations, we follow the IPA convention of indicating phonemes (lower-side symbols) between slashes. These slashes will not really appear in the output string.)

    braço brasu

  • The orthographical "ss" is always pronounced /s/. In this and following illustrations, the lexical and surface strings are lined up character pair by character pair, with the 0 (zero, also called epsilon) filling out the places where a lexical symbol maps to the empty string. These zeros will not appear in the final output.

    interesse interes0i

  • The orthographical "c" before "e" or "i" (or accented versions of these vowel letters) is always pronounced /s/.

    cimento simentu

  • The orthographical digraph "ch" is pronounced /$/.

    chato $0atu

  • Elsewhere (i.e. not ch), orthographical "c" is always pronounced /k/.

    casa kasa

    No "c" should appear in surface strings.
  • The orthographical digraph "lh" is realized as /L/.

    filho fiL0u

  • The orthographical digraph "nh" is realized as /N/.

    ninho niN0u

  • Elsewhere, "h" is silent and is simply realized as 0 (zero, the empty string).

    homem 0omem

  • The orthographical digraph "rr" is always realized as /R/. Also, the single r at the beginning of a word is always realized as /R/. Elsewhere, r:r, i.e. lexical "r" is realized as /r/.

    carro	rápido	caro	cantar
    kaR0u	Rapidu	karu	kantar

  • The unaccented "e" is pronounced /i/ at the end of a word, and when it appears in the context between "p" and "r" at the beginning of a word; e.g.

    peruca	case
    piruka	kazi

    Between the lexical "e" and the end of word there can be an optional "s". Elsewhere e:e.

    cases kazis

  • An "o" is pronounced /u/ at the end of a word.

    braço brasu

    Between the lexical "o" and the end of word there can be an optional "s". Elsewhere o:o.

    braços brasus

  • A single "s" is pronounced /z/ when it appears between two vowels.

    camisa	case
    kamiza	kazi

    Elsewhere s:s (but see above where s s -> s).
  • A word-final "z" is realized as /s/.

    vez ves

    Elsewhere, z:z.
  • A "d" is pronounced /J/ when it appears before a SURFACE sound /i/. (N.B. This change occurs in the environment of any SURFACE /i/, no matter what that surface /i/ may have been at the lexical level.) Elsewhere d:d.

    lisse	verdade	paredes
    Jis0i	verdaJi	pareJis

  • A "t" is pronounced /C/ when it appears before a surface sound /i/. (N.B. This change occurs in the environment of any SURFACE /i/, no matter what that surface /i/ may have been at the lexical level.) Elsewhere t:t.

    tio	partes
    Ciu	parCis

  • The vowels are a, e, i, o, u, á, é, í, ó, ú, ã, õ, â, ê, ô, ü and à. All lexical symbols map to themselves on the surface level by default.

Testing Portuguese Pronunciation

Write a set of that performs the mappings indicated. As in the kaNpat example, the rules should be organized in a cascade, with the composition operator (.o.) between the rules. Be very careful about ordering your rules correctly; the rules cannot be expressed in exactly the same order as the facts listed just above. Compile the rules using the read regex from utility in and test them using the apply down utility.

You should be able to handle the following examples, entering the lexical (top) string in each case and getting back the surface (bottom) string. (The zeros are not shown here and should not appear in your output.) To facilitate the testing, you can type all the input (upper-side) words into a file, called something like mydata, and tell apply down to read the various input strings from that file.

Be sure to test ALL the examples to make sure that your rules are really working as they should. Modify your rules and re-apply the input words until the grammar is working perfectly.