Language and Linguistics

ASHOK R. KELKAR

Bulletin
TREE DIAGRAMS, LABELLED BRACKETINGS, AND SUCH

As everybody who has any experience in the matter knows, the typing and printing of labeled bracketings and (even more so) tree diagrams is a headache. Writing or drawing them by hand on paper or on a blackboard is of course quite easy; but inserting a handwritten diagram in printing by using a block is quite expensive—particularly when one considers that the block may never be used again. Composing them has its difficulty; the result (as readers of IL have no doubt realized) often looks unattractive or space-consuming or both.

I propose certain conventions for printing tree diagrams and labeled bracketings that will make for ease of printing and low costs. The best way to do is through illustrations. Each example appears in four versions : (a) conventional tree diagram; (b) its proposed revision; (c) conventional labeled bracketing; (d) its proposed revision.

(1a) X (1b) X

A B C

A B C P L M S T U

P L M STU Q R

Q R

A possible variation of (1a), (1b) is one in which P, Q, R, S, T, U are all written in a horizontal line with the branches AP, LQ, MR, and the triangle SCU appropriately elongated. In (1b) the triangle will remain as it is; the bar over STU will be lowered. Note that the difficulty of ‘composing’ (1a) with all obliques slanted at 45 degrees is eliminated in (1b) and the (1b) is more space-saving than the conventional genealogical tree.

[P] [Q] [R] [STU]

(1c) A A L L M M C C

B B

X X

Here some ‘information’ is lost in passing from (1a) or (1b) to (1c) in that [STU] is not appropriately ambiguous as the triangle is between [ [S] [TU] ] and [ [ST] [U] ] and [ [S] [T] [U] ]; it will mean the last alternative

[ld] X [A [P] B [L [Q] M [R] ] C [ Δ STU] ]

The reading convention for (1d) is simple; it presupposes that every bracket (i.e. every node in the tree) is labeled. Sister-adjunction and Chomsky-adjunction present no problem. Consider sister-adjunction of C to B in (2).

(2a) X (2b) X

A C B

A C + B

(2c) x[ A C + B]x (2d) x [A C B]

And now Chomsky-adjunction of C to B in (3).

(3a) X (3b) X ; better X

A C # B A B*

A C # B

C B

(3c) x [A C # B] x (3d) x [A B* [CB] ]

Generative-semantics-style diagrams tend to be simple horizontally but with long vertical extension. Consider (4).

(4a) X (4b) X ; better X

A B

A B X

A B A B

X

X

A B

A B

(4c) A [A B] (4d) X [A B [X [ A B] ] ]

X X

X B B X

I have not separately considered the problems of typing; but it will be noticed that typing will be simpler too.

I have not introduced any plus marks or other boundary symbols in the proposed notations so far. Such symbols can actually be used for indicating the mode of overt ordering between the constituents :

(i) Significant order : [A + B] (i.e. A, then B and thus different from [B + A])

(ii) Nonsignificant order

(a) Rigid order : [A B] (i.e. A, then B, and never possibly B, then A) e.g. [quick⌒ ly]

(b) Fluid order : [A X B] (i.e. A, then B or B then A without any difference, at least any noticeable difference) e.g. Hindi (hai X nahiⁿ]

(c) Simultaneous order : [A, B] e.g. [Imp, Neg],

[ + COUNT, + ANIM], [ + HIGH, + BACK]

When no particular kind of order is to be indicated, no boundary symbol need be used : [A B].

Note how the comma ordering under (ii) (c) can be used to convey elaborate feature analysis at the level of phonology and syntax-semantics in an easy-to-print form. Thus, a vowel harmony rule calling for delayed vowel assimilation could be presented as :

[ V, + high] [α back, β round] / [V, α back, β round] (C)—

Some of the insights of prosodic analysis could be incorporated by adopting labeled bracketing. Thus :

I [ C [ + tense, + labial, + stop] ] F [ V [ + high, ^-- back, ^-- round], C [ + nasal, + velar] ] T [ + high, + fall]

could be offered as a phonological analysis of ping resolved into an initial element P---, a final element ing, and a tonal element high fall.

Halliday-style choice and chain diagrams read from left to right and tend to have great horizontal extension with great inconvenience in printing. They could be recast from top to bottom—the solid horizontal bar standing for chain and the wavy bar for choice. Thus, a corpus consisting of

X [A B C [P] ], X [A B C [Q] ], and X [A B C [R] ] can be generated by (5b) or (5d) below :

(5b) X

A B C

P Q R

P P

(5d) X [A B C [ Q ]]; better X [A B C [{ Q}] ]; even better X [A B C [{P Q R}]]

R R

Notice that the elongation of [ ], ( ), { } is not always necessary.

If one desires to introduce functional labels, the Tagmemicist device of using the ‘slot : filler’ coupling can be useful.

(6a) X (6b) X

a b

a : A b : B

A B

(6d) X [a : A b: B]

The Tagmemicist ± A is of course better represented as (+ A), so that

(7) (± ( ± A + B)) + C + (D + E)

will be recast as

(7d) ( + (A) + B) + C + [D + E]

May I express the hope that the stratificationists will also find their way to more printable diagrams on similar lines (pardon the pan)?

Before I conclude, let me point out that ʘ as a zero symbol is less confusing than the slashed 0 (compare the high-mid front rounded vowel and the voiceless bilabial frictive) and easier to print (very few presses have the slashed zero, most presses have the geometric symbol for a circle : The only merit of the slashed zero is of course that it is easy to type.

[Readers of IL are urged to try out these suggestions and earn the gratitude of editors, proofreaders, printers—perhaps even readers! They are also invited to contribute items to the new feature “Shop Talk”.—Ed.].

COLOPHON

This was published in Indian Linguistics 35:170-2, 1974.