^* This work was supported in part by the National Science Foundation.

¹ As well as to the 7090, except for timing. The 7090 is almost identical to the 709 except that it is transistorized and therefore about five times as fast as the latter. Although the 7090 is becoming very popular in industry, the 709 will doubtless remain more widely used in universities for some time, because it is available to them at very attractive prices.

I am grateful to Samuel E. Martin for several suggestions regarding the manner of presenting this material.

² 48 different possibilites may be used for input purposes, but only 47 are distinguished by the printer. The reason is that hyphen (binary 001 100; bottom of first column) and minus (binary 100 000; top of third column) are separate characters on the keypunch machine but are printed identically. Some machines made by other manufacturers accommodate larger numbers of characters.

³ Much of the work involved in designing this system was done by Alfred B. Hudson. In the table, the symbol @ represents BCD hyphen (cf. footnote 2). Since the only function of this BCD character in the system is to distinguish Cyrillic letters from Roman letters and Arabic numerals, the neutralization of hyphen and minus in output will be of no consequence for the ultimate use of the system, namely translation into English.

⁴ Thus the word of Figure 3 could have two additional meanings. As a fixed-point number with binary point to the right of position 35, it has a value of about 26 billion (2³⁴ + 2³³ + 2²⁸ + 2²⁷etc.). As a floating point number, its value is in the neighborhood of 30 quintillion.

⁵ For those readers who are already familiar with the 704, it may be explained that ZAC is an extended code which is recognized by some assemblers (including BC SAP) and assembled as PXD with a zero tag.

⁶ A cycle is 12 microseconds. Shifting instructions require two cycles for the first nine places of shift, one additional cycle for each additional 12 places or portion thereof. ANA takes three cycles, ANS four cycles, CAS three cycles (see Table 3). All other operations take two cycles except for multiplication, division, floating-point arithmetic, and input/ output, which take longer. For all operations in Table 3, 709 timing is the same as on the 704.

⁷ The additional arithmetic operations include Add Magnitude, Subtract Magnitude, Multiply, Multiply and Round, Round, Divide or Halt, Divide or Proceed, Add and Carry Logical, and a set of floating-point arithmetic operations. For other instructions and further details in general, see the reference manual for the machine in question, obtainable through local offices of the manufacturer or at university computing centers. On ZAC, see footnote 5; for timing of operations see footnote 6.

⁸ The operation codes for all operations are given in the reference manual (cf. footnote 7). Various ingenious programming devices, including additional methods of providing program branches, are possible through taking advantage of and performing operations upon the actual operation codes of instructions.

⁹ For a device using two index registers in an extremely rapid system for looking up dictionary entries see Sydney M. Lamb and William H. Jacobsen Jr., A High-Speed Large-Capacity Dictionary System, Mechanical translation, Vol. 6 (in press). In double (or triple) indexing, the logical OR (cf. Table 3, V-B) of the contents of the index registers constitutes the amount by which addresses are modified.

¹⁰ See A. Newell, F. M. Tonge, E. A. Feigenbaum, G. H. Mealy, N. Saber, B. F. Green Jr., and A. K. Wolf, Information processing language V manual: Section I, The elements of IPL programming, RAND Corporation Paper P-1897; Section II, Programmers reference manual, RAND Corporation Paper P-1918 (both Santa Monica, 1960).

¹¹ See Victor H. Yngve, A Programming Language for Mechanical Translation, Mechanical translation 5.25-41 (1958).

¹² The stress is on the first syllable. The name is, of course, based on the model of psycholinguistics and ethnolinguistics. Another possible name for the field is computational linguistics, but it has a difficulty in that to many people computational tends to suggest not just the computer but also, or instead, numerical calculation, so that computational linguistics sounds like something closely akin to statistical linguistics. Mechano- means machine but also suggests mechanical and mechanization, as these terms are often used without reference to any actual machinery, as when people refer to a ‘mechanical discovery procedure’, and that idea comes very close to the heart of the kind of thing which we are concerned with in this field.

¹³ See, for example, Paul L. Garvin, Automatic Linguistic Analysis—a Heuristic Problem, and Sydney M. Lamb, On the Mechanization of Syntactic Analysis, both to appear in the Proceedings of the International Conference on Machine Translation and Applied Language Analysis.

¹⁴ The term encode is here intended to have a somewhat different meaning from produce, since the process of providing target-language sentences in translation involves fewer operations than total production, as in a conversation machine. The translation process involves (1) recognition of the units of content, i.e. the sememes, which are represented by the sequences of letters or sounds (i.e. decoding) ; and (2) the furnishing, in the proper order, of target-language representations (in steps, first morphemic, ultimately graphetic or phonetic) for these sememes (i.e. encoding). Further details are given in my paper, MT Research at the University of California, Proceedings of the National Symposium on Machine Translation (Englewood Cliffs, 1961).

¹⁵ Problems which students have worked on in my seminar in mechanolinguistics have included the following : A program to generate the forms of the preterite of verbs in written French; a generator of inflected forms of English verbs; a program for partial comparative reconstruction, using vocabulary lists of two related languages; a program for partial phonemic analysis, using material in phonetic transcription as input data; a program to convert from a morphophonemic transcription of Arabic to written Arabic ; a model of the expansion and extinction of languages and linguistic families in large areas over long periods of time; a concordance maker; a program which provides organized lists of distribution class sequences found in texts; three programs to search texts for examples of different kinds of sequences; a ‘text mutator’, which revises the segmentation and assignment of distribution class symbols in a text to give effect to successively more refined analyses; a program to execute Sanskrit sandhi rules as stated by Panini; a generator of grammatical strings of lexemes; and a program which determines the syntactic structures of sentences. The last two programs are designed to operate for any language, provided they are given properly formulated syntactic information to make use of.