1. ‘Morris in Latvia’

Morris came to the United States in May 1940. He had escaped at the last minute the Russian invasion of his native Latvia, to be followed a year later by the German occupation of the Baltic states. That he and his immediate family made it out, on the last train out of Riga and then on the last US-bound ship out of Italy, was such a convergence of lucky accidents and so improbable an outcome that Morris would revisit his exodus in later years, in conversation with friends and in a 1984 book manuscript. The first part of this note follows the narrative of Halle Reference Halle1984, and borrows its title.

He was born in the town of Libau (Liepaja), on July 23 1923, the son of Lisa, née Kahan, and of Irving (Iska) Pinkowitz. His parents’ families came from small Lithuanian and Polish shtetls and had settled only recently in Latvia. Here is a portrait of one grandfather, by way of introduction to the atmosphere where he grew up:

He grew up in a world where it was normal to be fluent in German, Russian, and Latvian. Older family members spoke Yiddish, children would be taught Hebrew in school. Both of Morris’s parents were well-educated, partly in German institutions. Both had interrupted their studies during the First World War and later again, due to financial insecurity. Morris’s mother had studied comparative Romance and Germanic philology and literature. Later, she worked as a teacher of history and foreign languages in a Riga school, to help support the family. Morris’s father had first tried to keep the family’s flour mill in Libau afloat, before giving up and striking out on his own in Riga, as insurance agent and traveling salesman, and later as manager of a major textile business and chief financial advisor to its owner.

The extended family was impressive in size and stayed in close contact. Morris’s description of the family starts with the generation of his great-grandparents and records every one of their descendants, their occupation and successive places of residence, their temperament and offspring, as if from first-hand knowledge. Careers ranged from professional and business people, to a British countess, to chronically destitute Luftmenschen. The more exotic characters, all dutifully recorded, included one retired uncle from Libau, on whose office door a large painted sign announced audience hours that bore no relation to the uncle’s whereabouts:

After every individual record, in the last couple of generations, Morris notes whether they survived the war. Of those he left in Latvia, none did.

The family was secular and inclined to assimilate. On High Holy Days, young Morris would fast ‘partly as an act of solidarity with my fellow Jews, partly as demonstration of self-control, a quality highly prized by my mother’. After 1933, Morris and his sister, who had started out in the Riga German-language public schools, had to move to a Jewish municipal school founded by a group of ‘mildly leftist’ young teachers. They had no choice, but the school was good. ‘It was without doubt the best school I ever attended and it had a profound influence on my entire life. […] [T]eachers did not view themselves as drill sergeants but as mentors to younger people.’ As he described them in 1984, they must have reminded him of his older self, from the math teacher who spent hours teaching students to set up algebraic equations, to the ‘extremely tough’ world history teacher, who ‘had tremendous enthusiasm for his subject and got me to read literally dozens of books’, to the Bible teacher who carried himself like a boxer and would shout at students, in Hebrew, ‘you are an ignoramus squared!’.

During this period, Morris ‘identified as left-wing socialist’, marched with the socialists through Riga, spent time reading, hiking, and arguing with members of Hashomer Hatsair, after ‘flirting with joining a communist underground cell for a while’, and took the militantly atheist side in debate with a Hasidic rabbi.

In 1934, a rightist coup in Latvia persuaded Morris’s father that any risk they ran by emigrating compared favorably to the risk involved in staying put. The evidence bearing on this question was then so slim that the extended family refused to move. Morris’s father began alone the slow process of persuading the US government to let them in. It took four years to document that they would not become a burden upon the state. Visa applications could only come after that. In fall 1939, the Nazi attack on Poland brought the Red Army to Latvia. Life continued to look safe, until the German invasion of Norway and Denmark in April 1940. The US visas had just arrived. Rather than wait for direct passage to America, the Halles decided to get out of Latvia immediately, by train. After multiple detours, they boarded their ship in Naples. Belgium and the Netherlands fell, France was attacked, and Italy joined the war on the Nazi side while they sailed.

2. An American education

They landed in New York City in May 1940. That fall, Morris enrolled as a senior at George Washington High School, in Washington Heights, NYC. Close to a quarter of his classmates were European refugees, among them Sylvain Bromberger, a future close friend, MIT colleague, and collaborator. After graduation, Morris started engineering studies at CCNY. When the United States entered the war, he was drafted in fall 1943, trained at Fort RitchieFootnote ⁴ and sent to the European front in early 1945. It was too late to see combat but he was attached as translator to Military Intelligence, at a POW camp near Troyes and in Paris, where he discovered the virtues of multilingualism (Egré Reference Egré2008: Session 1).

Discharged from the army and back in the US in early 1946, Morris opted for a career as a Russian language specialist, initially less out of interest for Russian than out of concern for steady employment. The place to acquire the right credentials was the University of Chicago, where George Bobrinskoy, a Russian count, Sanskritist, and student of Franklin Edgerton, of Sievers-Edgerton Law fame, was in charge of the Russian program. In the end, the Sanskrit connection in Chicago worked against a career in Russian: Morris soon befriended the Indo-Europeanist Giuliano Bonfante, who wasted no time discovering that the right career for him was linguistics, and the right teacher Roman Jakobson, then in exile at Columbia. Morris finished an MA thesis on early Soviet debates about state control over literary works (Halle Reference Halle1948)Footnote ⁵, and then headed back to New York City, to look for Jakobson. But Jakobson, it turned out, was about to move to Harvard. As Morris tells it, of the sixteen students in Columbia’s Slavic program, ten followed Jakobson to Harvard. The other six dropped out (Egré Reference Egré2008: Session 1).

3. The context of early work

Morris’s first two years at Harvard must have been taken up with learning Russian philology and becoming acquainted with Jakobson’s and Trubetzkoy’s works on distinctive features and phonological universals. The linguistics scene was dominated then by Neo-Bloomfieldian linguists, who conceived of their work as having a sharply different orientation not only from the Prague School but also from Edward Sapir’s and, to an extent, Leonard Bloomfield’s work.Footnote ⁶ An in-depth survey of post-Bloomfieldian American structuralism can be found in Anderson Reference Anderson1985. The excerpts below provide just a sense of the climate of opinion in the American linguistics of the 40s and early 50s, as the backdrop against which one might read Morris’s early works. Much more along these lines can be found in Chomsky Reference Chomsky2021.

• ○ A linguist’s job is to classify patterns of behavior:

  ‘As linguists we do best to leave the physiological but also psychological or sociobiological problem to others […]. What is left to us is the specific form of a general problem of cultural anthropology: the development of operational techniques for deducing habits from behavior. With these strictures, the terms “language” and “speech” [Saussure’s langue and parole], can well enough be used: “speech” is behavior, “language” is habits.’ (Hockett Reference Hockett1942:99, emphasis mine)

• ○ Deducing linguistic patterns from principles, seeking to explain them, is not the way:

  ‘American Linguistics […] got its decisive direction when it was decided that an indigenous language could be described better without any preexistent scheme of what a language must be […]. From that time to today, the style of American Linguistics continues unbroken […]. The abandonment of deduction in favor of induction has never been reversed.’ (Joos Reference Joos1958:4, emphasis mine) ‘American linguistics owes a great debt to [the] stimulation [of the Prague Circle]; but in the long run those ideas were not found to add up to an adequate methodology. Trubetzkoy phonology [sic] tried to explain everything from articulatory acoustics and a minimum set of phonological laws taken as essentially valid for all languages alike, flatly contradicting the American (Boas) tradition that languages could differ from each other without limit and in unpredictable ways, and offering too much of a phonological explanation where a sober taxonomy would serve as well. Children want explanations, and there is a child in each of us; descriptivism makes a virtue of not pampering that child.’ (Joos Reference Joos1958:106; emphasis mine)

• ○ Levels of analysis must not be mixed

  ‘[P]honemics must be kept unmixed from all that lies on the opposite side of it: kept uninfluenced by the identities of the items of higher rank (morphemes and so on) which the phonemes “spell”, and hence free from all that their identities entail, such as their meaning and their grammar. […]. [T]he ghost of the slain dragon [i.e. mixing of levels] continued to plague the community of linguists under such names as “grammatical prerequisites to phonemic analysis” [Pike Reference Pike1947] and has not been completely exorcized to this day.’ (Joos Reference Joos1958:96). ‘It is not for us to say that the vowel alternation is more significant than the suffix for preterite marking of the verbs creep, keep, sleep, weep, or any such psychologistic formulation; it is enough to point out that the final alveolar stops in crept, kept, slept, wept were not correlated to the same set of micro-phonemes as the final stops of heaped, wrapped, helped.’ (Twadell Reference Twaddell1935, in Joos Reference Joos1958:79)

4. Early work

Morris’s first solo publications (Halle Reference Halle1951, Reference Halle1953a) offer analyses of inflectional allomorphy that seem unconcerned about the danger of mixing levels, in possible violation of the dicta above: all his phonological rules are conditioned by morphological entities like stem, desinence, aorist. The project in these papers was constructive, not polemical, as they aimed to show that the simplest accounts of verbal allomorphy, in Old Church Slavonic and in German, consist of selecting one basic root allomorph from which the rest of the verbal paradigms can be generated by rules. Both papers argue that the simplest derivation of the paradigm is obtained if this basic allomorph is the one that preserves the most conjugation-specific distinctions.Footnote ⁷ The theoretical model is that of Jakobson (Reference Jakobson1948), who had argued earlier for a similar choice of base in Russian, in almost generative terms.Footnote ⁸

The immediately following works—including Jakobson et al. Reference Jakobson, Fant and Halle1952, Cherry et al. Reference Cherry, Halle and Jakobson1953, Halle Reference Halle and Mueller1954, Reference Halle1956—show Morris at work as a phonetician. He had been hired in 1951 at MIT as instructor in German, Russian, and French. He also joined the Research Laboratory of Electronics (RLE), the continuation of MIT’s wartime Radiation Lab (RadLab)Footnote ⁹, turned after the war into a conglomerate of projects that included ‘information theory, cybernetics, audition, psychoacoustics and speech perception’ (Halle Reference Halle1996, Harris Reference Harris2011 [1974]).Footnote ¹⁰ The members of RLE’s Speech Analysis group included at the time Ken Stevens and Gunnar Fant. Morris was pursuing the Jakobson-initiated project of identifying acoustic correlates for distinctive features.

The collaborative pieces from this early period read like Jakobson. The reader may get a sense of what that means from Chomsky’s (Reference Chomsky1957) review of Jakobson & Halle Reference Jakobson and Halle1956, which contains passages like this: “Both essays are written in a rather picturesque and inexplicit style which, to me at least, presents a bar to comprehension. It is difficult to determine which statements are empirical hypotheses and which are true by definition.’ (ChomskyReference Chomsky1957:334)

If Morris’s own voice and preoccupations can be recognized today, it’s in the more obscure solo writings of this period. In the passage below, he criticizes the authors of a report on X-ray tracking of articulatory movements for focusing only on variability between utterances while failing to address the complementary and essential question of invariance:

Morris’s students will remember similarly stern admonitions, from ten to sixty years later, about missing the whole point of scientific inquiry.

In a review from the same period, Morris takes to task another linguist for failing to extract the signal from the noise. Charles Hockett had invited the readers of his Manual (Hockett Reference Hockett1955:210ff) to visualize the relation between a string of phonemes and the corresponding acoustic signal as an orderly row of Easter eggs being passed on a belt, through a wringer. At the other end of the wringer, an inspector waits to identify the eggs, but fails: the eggs were squashed. The inspector is the listener. That was Hockett’s vivid metaphor of speech production and perception, to which Morris reacted as follows:

5. Decoding the message, and phonetics

Actually, Hockett was defending in his Easter egg passage an approach to speech perception in which listeners entertain the possibility of extensive articulatory overlap in spans of continuous speech. Taking overlap into account means that candidate phonetic categories for any slice of time must include two or more superimposed phones. Learners use trial and error to reconstruct in such signals specific combinations of non-overlapped phones. An element of abstraction—the properties of the phone in isolation—is essential to Hockett’s idea. But it’s also true that any speaker knowledge about the rules governing speech production plays no role in this approach to speech perception. Morris had in mind a different approach, based on ideas he was developing at the time: cues are cues to feature values, not to holistic segments; cues may have to be evaluated relative to context rather than representing context-invariant values; and, critically, listeners use analysis by synthesis, their knowledge of the finite set of processes active in the language that could have resulted in the specific pattern of overlap observed, given a certain input. The code, then, is the phonetic grammar.

Morris’s ideas about production and perception were informed by his phonetic work, which includes collaborative studies on the acoustic properties and perceptual cues to the stop/fricative and place contrasts in stops (Hughes & Halle Reference Hughes and Halle1956, Halle et al. Reference Halle, Hughes and Radley1957). It is instructive to read these works next to a short contemporaneous piece, Halle Reference Halle and Mueller1954, which deals with the more general question of how and why linguists practice measurement and analysis. This squib was printed in the proceedings of a roundtable on spectrographic analysis, but it seems intended as a first manifesto against the ‘separation of levels’, the idea dominant in the 1950s, and illustrated by some of the Joos quotes above, that phonemic analysis must proceed without regard to higher-level syntactic or semantic considerations, and without considering lower-level phonetic information.Footnote ¹¹ Morris argues in the 1954 piece that phonetic measurements and phonemic analysis are meaningful only when pursued as complementary sides of a unified enterprise, the study of spoken language as a means of transmitting information between speakers. The phonetician selects specific units to measure because they have the potential to function as carriers of information. The phonemicist assigns variants to the same phonemic category based, in part, on phonetic considerations: their acoustic or articulatory similarity, as itemized by shared distinctive features. These activities of measurement and analysis define a joint project, sharply distinct from the engineer’s project to recreate digital copies of speech tokens, and they can only be pursued in tandem:

6. Sound Pattern of Russian

The last sentence in the Halle Reference Halle and Mueller1954 fragment above—parts of an analysis can be evaluated only against the grammar ‘in its entirety’—may look obscure in the local context, but its message returns in fuller form in The sound pattern of Russian (SPR; Halle Reference Halle1959a), a revised version of parts of Morris’s 1955 Harvard thesis.

The initial project in SPR had been to compare the results of a distinctive-feature analysis of Russian phonology with the evidence provided by the acoustic analysis of representative Russian utterances. What questions would one hope to answer by this procedure? The answer could be that SPR uses a feature set whose acoustic correlates had been provisionally defined in Preliminaries on the basis of English data alone, so its ability to extend to Russian, or anything other than English, was an open question. Further, the grammar proposed in SPR identifies contexts of neutralization and contexts of intact realization, for each feature. The acoustic analysis promised to test the predictions made by this analysis. As an example, Tables V-1,2 (Halle Reference Halle1959a:127ff) report on the success of the joint predictions of the vowel reduction rules, which alter vowel height, plus the acoustic correlates assigned to the features [±diffuse] and [±compact]—later redefined as [±high] and [±low]—when confronted with the formant measurements of spoken Russian. (The predictions fared rather well, except in contexts where the unaccented vowels fluctuate in height, between [ə] and [i].) Passages like this show how the author intended to redeem the book’s promise that acoustic data will be brought to bear on phonological matters. In today’s terms, SPR was, among other things, a first laboratory phonology project, an effort to identify and test the phonetic predictions of a phonological analysis.

SPR is better known today as the first published application to phonology of the idea of an evaluation metric, the standard guiding the choice between alternative analyses (Chomsky Reference Chomsky1957:51ffFootnote ¹²). The early version found in SPR consists of one condition of adequacy, condition 5 (Halle Reference Halle1959a:29–30), which compresses the grammar by eliminating predictable feature specifications from lexical entries. Its consequence is the first appearance of systematic underspecification in a phonological analysis. In the text of SPR, a functional consideration is offered in support of featurally sparse lexical entries—they are said to speed up the selection of target morphemes in speech production (Halle Reference Halle1959a:29)—but the more easily defended point is that underspecification is the direct consequence of the drive for a simpler grammar, which condition 5 implements together with a hypothesis about feature hierarchies.

The global impression is that a few broad principles dictate all analytical decisions in SPR and that the inspection of the Russian detail is undertaken to illustrate these broad principles. In this context, we understand the end of a passage cited earlier: as analysts, we ‘evaluate all our statements by their effects on the description of language in its entirety’. Local decisions—for example, which feature value to specify underlyingly and which other value to predict by rule—will affect the size of the overall lexicon and the applicability of phonological rules. Their effect, then, can be assessed only in the context of the full analysis. SPR’s analysis of Russian phonology is indeed complete, according to Ferguson’s (Reference Ferguson1962) extensive and critical review. Then this is one of the few analyses of one phonological system that aimed not just for the occasional insight but for full coverage, the only way in which condition 5 could be evaluated.

Several proposals introduced in SPR inspired later work. SPR advocates filling in underspecified lexical entries by morpheme structure rules (MSRs), and then by phonological (P) rules. The MSRs apply before but also during the phonological derivation, to avoid artificially splitting a unitary process into an MSR and a P rule (Halle Reference Halle1959a:61, Section 2.6). This detail, and the systematic elimination of redundant detail from lexical entries evokes aspects of the much later work on radical underspecification (Kiparsky Reference Kiparsky, van der Hulst and Smith1982, Reference Kiparsky1985, Archangeli Reference Archangeli1984), a theory that allows P-rules to access drastically underspecified lexical entries.

An early version of what is now known as the contrastive hierarchy (Dresher Reference Dresher2009) appears in Jakobson & Halle Reference Jakobson and Halle1956 and in SPR’s discussion of context-free redundancy rules (Halle Reference Halle1959a:34ff). See Dresher’s (Reference Dresher2009:76ff) extensive discussion of the use and history of contrast hierarchies, including in SPR. A feature hierarchy, with major-class features providing the first partitions, guides the order in which featural blanks are assigned. This order dictates the choice between alternative ways of enforcing underspecification. In Halle Reference Halle1959b:513ff, a feature hierarchy is shown to predict the distinction between neutral vowels in a harmony system and the vowels that actively participate in harmony, by triggering and blocking it. The example used there is that of the Finnish vowels [e] and [i], which mostly don’t trigger, and never block, the Finnish front/back harmony. [e] and [i] are also the vowels whose front values are predicted to be underlyingly unspecified by a feature hierarchy which, when translated in post-Jakobsonian features, ranks [±round] > [±low] > [±high] > [±back] (Halle Reference Halle1959b:514). If this ranking is otherwise justified, this is a very significant finding. This argument is Morris’s pre-emptive response to later criticism that no empirical connection can be established between rule generality and the requirement that lexical specifications should be minimized (Anderson Reference Anderson1985:296). In Finnish, the connection is this: the most general formulation of Finnish front harmony, in which every segment that possesses a [±back] value participates in harmony, is possible only if features are minimized in the lexicon.

SPR opens by asking what it means for two expressions to have phonologically identical representations. As seen earlier, Morris saw this as a fundamental question. The answer offered in SPR is that linguistically significant identities are defined in underlying representations. The properties of URs, in turn, are identified in a search that combines phonological, morphological and semantic information. This is an explicit break with the neo-Bloomfieldian practice of separating levels of analysis. Morris’s empirical arguments against that practice, such as those based on obtaining unified analyses of Russian voicing assimilation (Halle Reference Halle1959b:22ff) and vowel reduction (p. 41), have received considerable attention elsewhere (Anderson Reference Anderson1985:32, Dresher & Hall Reference Dresher and Hall2021).

7. Analysis by Synthesis: Reverse Engineering in Perception

Some of Morris’s projects from the 50s and 60s go considerably beyond phonetics and phonology. They are connected to the rest of his linguistic work by his abiding interest in reverse engineering covert mechanisms.

In 1959, Halle and Stevens offered a first domain-general characterization of the method by which a receiver reconstructs the underlying message from a coded signal, equipped with just knowledge of the coding function. This is the first analysis by synthesis (A-by-S) paper, a report that builds on several projects Morris was engaged in at the time: automatic speech recognition (Hughes & Halle Reference Hughes and Halle1960, Halle & Stevens Reference Halle and Stevens2002 [1962]), an automated morphological parser (described in Halle Reference Halle1960), and a reader of cursive script (Eden & Halle Reference Eden, Halle and Cherry1961).

An A-by-S decoder is in possession of the function that has encoded an underlying message and uses his knowledge of this function for the purpose of verifying the identity of the message.Footnote ¹³ In speech perception, an A-by-S parser exploits knowledge of a production grammar to construct candidate lexical entries from which the grammar could have generated the perceived sound stream. The conclusion of Halle & Stevens Reference Halle and Stevens1959 offers the following comparison of the A-by-S scheme with alternatives entertained at the time, including the first version of the motor theory of speech perception:

The Eden and Halle (Reference Eden, Halle and Cherry1961) work on cursive decoding, unknown today, is a set of prolegomena to a generative grammar of handwriting. It uses as the coding function a grammar that assembles cursive letters from four elementary strokes, plus rules that link up and compress strokes into letters, and then letters into connected strings. Its objective is to deliver a parsing algorithm, but also same/different judgments, much like a grammar of speech:

More distantly related to the decoding work, and still relevant today, are Morris’s ideas about spelling conventions, which ranged from the suggestion that English spelling is a deeply rational system, despite opaque appearances (Halle Reference Halle, Goodman and Fleming1968, Chomsky & Halle Reference Chomsky and Halle1968:49ff, 184 fn. 19)—a sentiment later partly endorsed by McCawley (Reference McCawley1974:51) and Burzio (Reference Burzio1994:52ff)—to the equally striking claim that Russian spelling, while transparent, is fundamentally irrational, because it fails to provide unified representations for categories that are identical underlyingly or identical throughout (Halle Reference Halle, Lapidus and Shevtsova1963).

8. Friends, colleagues, students, and family

Morris had landed at MIT in 1951, shortly before Noam Chomsky arrived from Penn at Harvard’s Society of Fellows. They came in contact that fall, when Carol Chomsky, then a twenty-one-year-old phonetician interested in Jakobsonian features, applied for a research position in RLE. She was interviewed for the job by someone who she thought was ‘an elderly European professor of very stern demeanor’. Morris protests: ‘I was twenty-eight at the time, and I could not have been too stern.’ (Halle Reference Halle1996). He met Noam Chomsky the same day and they had the first of many arguments (Chomsky, in Liberman Reference Liberman2016). Morris recounts as a first topic of debate his own insistence that the object of linguistics—what one learns, when one learns a language—had to be kept distinct from how one learns it. That second thing is psychology (Egré Reference Egré2008:Session 2). They reached agreement, and became friends and collaborators.

When Chomsky’s Junior Fellowship ended in 1955, he applied for a position with MIT’s Department of Modern Languages, and Morris recommended him. In an attempt to verify his information, the head of the department traveled by subway to Widener Library and interviewed Chomsky in the stacks. The interview was satisfactory, Chomsky was hired as foreign language instructor, and joined the RLE as a member of the Mechanical Translation group. Soon, he was teaching introductory courses in linguistics and attracting students and other linguists to MIT, including G. Hubert Matthews, Edward Klima, Jerry Fodor, Jerrold Katz and, eventually, Jakobson.

The first Ph.D.s were awarded for theses in linguistics in 1960, in MIT’s Department of Electrical Engineering, to which RLE was affiliated. They were a study of English nominalizations, by Robert Lees, and a report on automatic speech recognition, by George Hughes. In 1961, an independent graduate program in Linguistics was created. All its members joined the RLE. The Quarterly Progress Report (QPR) of RLE started publishing at regular intervals notes on pre-Islamic meters, instrumental adverbs, and proto-Slavic diphthongs, alongside reports on stroboscopic research, AI, and neurophysiology. In the beginning, QPR notes on linguistics were prefaced by the group’s statement of research objectives, which emphasized the interplay between empirical study and theory, and its interdisciplinary connections.

The leader of the new department, whether bearing an official title or not, was Morris. In a history of the early years of linguistics at MIT, Harris (Reference Harris2011 [1974]:13) writes: ‘[Halle] has been its chief architect and builder. […] Intellectually, spiritually, and practically, [he] has been the strongest single force in building the group and in guiding and guarding its growth.’Footnote ¹⁴ Morris initiated the distinctive properties of graduate education in linguistics, which later spread from MIT to other programs: a curriculum that equipped students to begin independent research immediately, rather than slowly gain mastery over established doctrines; the practice of collective advising; the insistence on funding students centrally and equally, rather than relying on individual research grants, to ward off the specter of PI-imposed restrictions on research; and the conditions that would allow students to study in the department, to encourage them to learn from each other.Footnote ¹⁵

The goal of these policies was to create independent thinkers. Personal example and sternly delivered advice were as important. Both were used to get students to present their ideas by laying out from the very beginning fully explicit analyses whose mechanical workings would be spelled out and illustrated. The very next admonition students got from Morris was to disbelieve their teachers. This has become known as the injunction ‘Argue with me!’ (Aronoff & Oehrle Reference Aronoff and Oehrle1984:xi, Liberman Reference Liberman2016: passim). The student who found nothing to criticize, especially in Sound pattern of English, was suspect (Dell, in Liberman Reference Liberman2016:4). Poring over published work was discouraged. Try analysis first:

If the student persisted in having bibliographic concerns, Morris would start muttering about ‘asses laden with books’. Without tangible analysis, overly scholarly stuff was branded ‘a garden of horrors’ (Wagner Reference Wagner2018). That said, while citation was optional, failure to acknowledge relevant evidence would elicit sarcastic remarks: ‘You know, science is supposed to be cumulative.’ (Schütze Reference Schütze2018). Still, those who engaged in empirical overkill would be admonished that ‘data is overrated’ (Fabb Reference Fabb2018). Those tempted to wax prematurely theoretical got their comeuppance too: ‘“First, you do the analysis,” he grumbled. “After, you get to argue about the theory.”’ (Harbour Reference Harbour2018). Those who followed the rules but still got nowhere were rewarded with the tough-love consolation of ‘good try but no cigar’ (Farmer Reference Farmer2018), or a Biblical bit of wisdom, ‘the wages of sin is death’, delivered with an encouraging chuckle. And those who built arguments on appeals to authority? This is for them:

There was a family, too, apart from the students. In 1956 Morris had married Rosamond Thaxter Strong, a Massachusetts native, graduate of the Massachusetts College of Art and Design, an artist and activist. They had three children, David, John, and Timothy. Their marriage lasted fifty-six years, to her death in 2011.

9. The Sound Pattern of English

The creation of MIT’s graduate program in linguistics in 1961 meant that Morris would teach not acoustics and foreign languages to engineers, but phonology, morphology and, very frequently, syntax, to future linguists (Harris Reference Harris2011 [1974]). The first major publication that followed this change in teaching duties was the The sound pattern of English (Chomsky & Halle Reference Chomsky and Halle1968; SPE), a work shaped by its authors’ interactions with their students. The sound pattern part in the title was meant to remind the reader of SPR, but also of Sapir’s (Reference Sapir1925) Sound patterns in language, which Morris had identified as a key influence on SPR (Halle 1959:13):

What was the one thing that SPE was about? What ideas of Sapir’s was it building on? The descriptive focus of SPE’s major analytical parts, Chapters 2–5, is on the phonological processes of English that derive stress and vowel quality, the latter being both a conditioning factor and a consequence of stress. Consonantal processes tend to be analyzed in SPE only if they interact opaquely with other processes. For example, the interacting pair of velar softening and s-voicing are carefully analyzed (Chomsky & Halle Reference Chomsky and Halle1968:228ff), possibly because vowel shift counterfeeds and counterbleeds velar softening, which in turn counterfeeds s-voicing. By contrast, the relatively transparent and vastly more productive rule of flapping is mentioned once in passing, only because it verifies the existence of stress differences that the system predicts (p. 191).

Such observations suggest that the choice of phenomena to analyze in SPE was deliberately oriented towards sound regularities that can’t be detected in surface distributions. The partial, surface predictability of English stress had been noted by Newman (Reference Newman1946, Reference Newman1948). But the focus in SPE was different: it was on the claim of complete predictability of stress and vowel reduction at an intermediate stage. To observe these regularities, layers of ordered processes had to be peeled off. The abstract representations thus uncovered reveal, the authors argued, an underlying pattern of pristine simplicity: a fully predictable stress that is fundamentally identical to that of an undeniably regular system, like that of Latin (Chomsky & Halle Reference Chomsky and Halle1968:29).Footnote ¹⁷

The reference to Sapir as predecessor makes sense in this light. One argument in Sapir’s Sound patterns is that certain regularities emerge only in abstract representations, when they are stripped of the effects of certain rules: ‘… such conditional variations as we have discussed once cleared out of the way, we arrive at the genuine pattern of speech sounds’ (Sapir Reference Sapir1925:42–43, emphasis mine). SPE argued similarly that the regularity of English stress and vowel alternations can be ascertained only when the abundant surface noise of English has been cleaned up. In possibly weaker form, this result has endured and continues to lend credibility to aspects of the grammatical architecture that allowed it to emerge in SPE. Those interested today in non-serial, constraint-based frameworks may find it challenging to analyze productive-yet-opaque interactions like that between English stress, final tensing, and flapping, among others.

In the service of the pristine pattern and the simpler rule, SPE got carried away sometimes. It notoriously derived righteous [raɪʧəs] from /rixt-i-ɔs/, to explain the orthographic <gh>. (Recall the authors’ belief in the fundamentally rational nature of English spelling.) The same /rixt-i-ɔs / → [raɪʧəs] mapping was invoked to explain why spirantization (t→s/_j) doesn’t apply, to yield *[raɪʃəs] (Chomsky & Halle Reference Chomsky and Halle1968:234): there’s an invisible fricative that blocks spirantization, that’s why. That fricative must be [x], to explain why it must later vanish. With this result acquired, more uses for [x] emerge on the same page of SPE, all reinforcing the merit of the [raɪʧəs] < /rixt-i-ɔs/ derivation: [x] explains the failure of trisyllabic shortening (TSS) in nightingale—its [aɪ] comes from [x] deletion, counterfeeding TSS—as well as the prevocalic engma in items like dinghy [dɪŋi], from /dinxi/. A list of other such passages was drawn up in McCawley Reference McCawley1974.

Today’s reader can’t tell if SPE’s drive to achieve perfect regularity, with the occasional ad absurdum consequence, was an exploratory strategy, or if it was rooted in a deeper intuition about phonological structure and learning. Morris showed an enduring fascination for the descriptive uses of what later became known, and rejected, as absolute neutralization (Kiparsky Reference Kiparsky and Fujimura1973 [1968]): underlying segments that never surface and only serve to alter the distribution of those that do. In later work, Morris cited approvingly the right < /rixt/ derivation (Halle & Mohanan Reference Halle and Mohanan1985) and his last paper appeals to [ɨ], an invisible 6th vowel of Latin, brought in to impose order on the Latin 3rd conjugation (Halle Reference Halle2019).

SPE’s second half outlined the grammatical architecture that made the English results possible, and supported it with data beyond English. It defended the choice of features and boundaries, the rule application conventions, including the cycle—an idea whose applications to syntax came after its use in phonology—and the system of abbreviatory devices that collapses multiple rules into one. By 1968, many of these ideas had appeared elsewhereFootnote ¹⁸ but their post-68 influence on analysis and theorizing must have been due to their role in formalizing the analytical results SPE obtained in English phonology.

The delight Morris took in discoveries like those in SPE was always tempered by his anticipation that the next bit of progress would inevitably upturn the order he worked to bring out of chaos. This is a topic in his address to the LSA as its President (Halle Reference Halle1975), but it’s best known through the admission that comes at the end of SPE, in Chapter 9. Here is a part of that passage:

After this paragraph, with the book essentially completed and the confession of impasse made, the authors proceed to sketch where a solution might be found: a theory of markedness. The book ends thirty pages later, on a technical note, ready for the next writer to pick up where they left off.

10. Post-68 innovations

The history of phonology records vigorous critical reactions to SPE. The immediate ones include the rethinking of English monomorphemic stress in Ross Reference Ross and Brame1972; McCawley’s massive 1974 review, supportive on the whole but sharply disagreeing on specifics; and Stampe’s dissertation (1979 [1973]), which argued for a different theory of markedness, against SPE’s abbreviatory conventions, indeed against any such conventions at all, against several rule application conventions (e.g. the one-rule-per-cycle condition) but in favor of the idea that true processes are persistent.Footnote ¹⁹ A book of commentaries (Goyvaerts & Pullum Reference Goyvaerts and Pullum1975) collected some of these criticisms, and others. One thought on reading all these reactions today is that many of the constructive and enduring objections were formulated by scholars whose critical skills had been honed through interactions with SPE’s authors.

It comes as no surprise to those who knew Morris that he proceeded after 1968 to embrace many deviations from SPE and actively participated in developing several, just as he vigorously fought back against quite a few. McCawley (Reference McCawley1974:61 and passim) had argued that reference to syllables and moras will simplify SPE’s stress assignment rules, a point reinforced by Hooper [Bybee] (Reference Hooper [Bybee]1972), Vennemann (Reference Vennemann1972), Stampe (Reference Stampe1979 [1973]:56ff), and in Kahn’s (Reference Kahn1976) syllable-based study of segmental processes. SPE had used instead units it referred to as strong and weak clusters, corresponding roughly to longer and shorter VC₀ intervals (Chomsky & Halle Reference Chomsky and Halle1968:29). After 1976, syllables were explored in Morris’s classes at MIT and incorporated into his work (e.g. in Halle & Vergnaud Reference Halle and Vergnaud1978).

SPE’s representation of stress as an n-ary feature underwent major revisions, in a section of Liberman’s (Reference Liberman1975) dissertation, in Liberman and Prince’s (Reference Liberman and Prince1977) study, and work inspired by it. Liberman Reference Liberman1975 records Morris as a participant in the exploratory phases of the metrical work. He adopted the metrical theory of stress, and later argued for a new version based on bracketed grids (Halle & Vergnaud Reference Halle and Vergnaud1987; cf. Hammond Reference Hammond1984). In subsequent work with Bill Idsardi, he came to favor instead metrical spans defined by unpaired parentheses (Halle & Idsardi Reference Halle, Idsardi and Goldsmith1995; cf. Idsardi Reference Idsardi1992).

SPE’s conception of phonological features as classificatory entities (Chomsky & Halle Reference Chomsky and Halle1968:65) gradually gave way to representations in which some features came close to representing not timeless attributes, but event-like entities related by association, a form of temporal overlap, to other such units. A series of dissertations advised by Morris—Woo Reference Woo1969, Leben Reference Leben1973, Goldsmith Reference Goldsmith1976, Sagey Reference Sagey1986—and related contemporaneous work by Anderson (Reference Anderson1976) and Clements (Reference Clements1980 [1976])Footnote ²⁰ took steps towards this conception of features. Morris adopted early some of these ideas (Halle 1970, Halle & Vergnaud Reference Halle and Vergnaud1978) and continued to explore them in later work (Halle et al. Reference Halle, Vaux and Wolfe2000).

The elsewhere condition (Kiparsky Reference Kiparsky, Anderson and Kiparsky1973) replaced SPE’s tentative identification of disjunctive ordering with the use of certain abbreviatory parentheses. Morris thought this was a major conceptual breakthrough (e.g. Halle Reference Halle1995) and later incorporated it as a central mechanism in distributed morphology (Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993, Reference Halle and Marantz1994).

Other innovations had more limited echoes. Among them is a new conception of cyclic stress, presented in Halle & Vergnaud Reference Halle and Vergnaud1987. The most striking proposal in that book extends the idea of distinct autosegmental tiers for distinct morphemes (as in McCarthy Reference McCarthy1979) to cyclically assigned stress. On each cycle, stress rules apply to create certain accentual structures. A copy of each cyclically created output is kept on a satellite stress plane, but the actual input to the next stress cycle is a deforested, stress-free tabula rasa.Footnote ²¹ The stress rules may apply again on one or more subsequent cycles, and again their output is separately recorded on another satellite stress plane. In a postcyclic component, the information preserved in this way from multiple intermediate cyclic outputs is—on a case-by-case basis—copied onto the main stress plane, and that gives it a chance to surface. Thus, the secondary stress in perìpherálity is said to have been copied, rather than inherited, from the satellite containing the cyclic output of perípheral. How one selects which satellite stress should be copied—e.g. why the secondary stress in solìdificátion is copied from solídify and not from sólid—remained less clear. But while the hypothesis of satellite stress planes left unanswered questions, the idea of copying stress from a related form, rather than inheriting it from a form embedded within the target form, reads today as related to the trans-derivational cycle that had been tentatively proposed by Chung (Reference Chung1983) and was later fully developed in optimality theory by Benua (Reference Benua1997) and others.

11. Morphology

SPE had proposed an unworkable theory of exceptions (McCawley Reference McCawley1974:63ff), an understandable failing for a project intent on detecting regularities behind the apparent exceptions, rather than on dutifully recording the latter. Morris kept working on this topic. The result were his later writings on morphology, beginning with the Prolegomena to a theory of word formation (Halle Reference Halle1973). This offered a single mechanism to model varieties of lexical irregularities in complex words: (i) non-compositional semantics (as in divinity vs. compositional serenity); (ii) idiosyncratic paradigm gaps (arrival vs. *derival); and (iii) idiosyncratic application of phonological rules (serĕnity vs. obēsity). The unified mechanism was a filter operating on the output of word formation, representing the input to lexical insertion, syntax, and later phonology. The filter would insert idiosyncratic lexico-semantic features; or instructions to block lexical insertion, as in *derival; or instructions to block rules, as with trisyllabic laxing blocked in obēsity. One notable feature of this model is the distinction between an initial list of atomic morphemes and a derived dictionary of full words, representing a subset of the potential outputs of word formation. The dictionary records actual words, those marked to undergo lexical insertion. The idea of a productively derived dictionary previews points later explored in lexical phonology (Kiparsky Reference Kiparsky, van der Hulst and Smith1982), with the difference that derived dictionary entries in Prolegomena have undergone word formation but not yet phonology.

In later work on morphology (e.g. Embick & Halle Reference Embick, Halle, Geerts, van Ginneken and Jacobs2005), Morris maintained his refusal to cope with lexical irregularity by listing complex expressions in the initial lexicon. No dual route for him. Other aspects of the Prolegomena filter-cum-dictionary model were abandoned. In the early 1990s, in collaboration with Alec Marantz, he combined earlier ideas and new ones into the framework of distributed morphology (DM). From SPR came the practice of inserting lexical items gradually into syntactic trees, with grammatical formatives represented only as syntactic features in early stages of the derivation.Footnote ²² From Anderson’s Amorphous morphology came the rejection of the classical morpheme, that is, the ‘one-to-one relation between components of meaning and components of form’ (Anderson Reference Anderson1992:70). The core of DM is the idea that, while each function morph is restricted to some morphosyntactic context, the relation between morphs and syntactic loci of insertion is flexible. The context of insertion may be underspecified, or completely unspecified, with the consequence that multiple morphs can compete for insertion at one site. The best available choice of morph in one context might not be the best in others. The syntactic nodes under which morphs are inserted can lose some of their features (through impoverishment) or can split (fission) pre-insertion, yielding analyses for varieties of deviations from the ideal one-to-one correspondence between syntactic function and morphemic identity. The last of the three festschrifts for Morris et al. Reference Matushansky and Marantz2013, reflects the rich body of work that the ideas in DM have generated.

12. Meter

In the late 60s, Morris Halle and Jay Keyser began a series of works on meter. The first of these, Halle & Keyser Reference Halle and Keyser1966, dealt mainly with Chaucer’s poetry, but was drawing on a general theory of poetic meter, presented there first. The theory defined the abstract structure of lines and possible deviations from such structure, the range of linguistic utterances that can fill out an abstract line template, and possible deviations from such regular mappings.

The goals of this and later studies in what became known as the field of generative metrics was parallel to those of contemporary generative works in syntax and phonology. Halle and Keyser were proposing models of the knowledge that underlies judgments of poetic wellformedness and relative complexity. As with syntax and phonology, the goal of generative metrics, as Halle and Keyser saw it, was not to model the data of some corpus but the knowledge internalized by the fluent participant who had been exposed to a representative sample from a poetic tradition. As a result, appeals to native speaker judgments played a role. As one example, Halle and Keyser (Reference Halle and Keyser1971:139) pointed out that 1a below is a fully-lawful English iambic pentameter, that 1b is quite impossible as an iambic pentameter, and that 1c is lawful but ‘more complex’ than 1a.

1. (1)

   1. a. The curfew tolls the knell of parting day

   2. b. Ode to the West Wind by Percy Bysshe Shelley

   3. c. O wild West Wind, thou breath of Autumn’s being

Spelling out what made the difference between ‘more complex’ and ‘quite impossible’ turned out to be a major undertaking. Halle and Keyser’s initial solution built on the relational notion of stress maximum (Halle & Keyser Reference Halle and Keyser1966:197), a syllable more prominent than both syllables it is adjacent to in a line, a peak surrounded by troughs. Line 1b above is impossible because it has misplaced in a weak position the stress maximum of by Pércy.

The stress maximum idea yielded two striking results (Halle & Keyser Reference Halle and Keyser1966:198–99). First, in iambic verse, an initial stress can land in weak position (Sílent upon a peak in Darien), because it’s not a stress maximum there. Second, corresponding mismatches of stress to ictus are impossible at the beginning of trochaic lines. The next fifty years of research brought refinements to the key notion of stress maximum (Kiparsky Reference Kiparsky1977), an articulated internal structure for the verse line (Prince Reference Prince, Kiparsky and Youmans1989), and a change in how metrical complexity is conceived of and computed (Hayes et al. Reference Hayes, Wilson and Shisko2012), among other results. Later contributions by Morris to the generative metrics project include Halle & Keyser Reference Halle, Keyser, Muraki and Iwamoto1999 and Fabb & Halle Reference Fabb and Halle2008, both of them cross-linguistic studies that experiment with the idea of meter as a mechanism whose function is to determine the length, not necessarily the rhythm, of a line.

The enterprise of generative metrics did not emerge in a vacuum. Jakobson Reference Jakobson and Sebeok1960 provides a glimpse into the discussions of poetic meter that Morris must have participated in during his student days in the 1950s. We find in Jakobson’s essay the idea of tacit poetic competence, the distinction between abstract verse design and verse instance; also allusions to the violable rules that link the rhythmic structure of a text to the abstract verse template, plus informal description of complex patterns of verse design, yet to be analyzed. What we don’t find are explicit proposals for actual grammars of verse, whose predictions could be confronted with the evidence of attested lines or of metrical intuition, and whose successive refinement could lead to further, unanticipated discoveries. We find all this for the first time in Halle and Keyser’s writings.

13. Constraint-based phonology

Morris’s interest in reverse engineering complex systems was mentioned earlier. He was most intrigued by mechanisms that conceal their input in successive layers of change. In his LSA address he briefly offered to the audience the idea of language as a ‘playful activity, a kind of game’ (Halle Reference Halle1975:527), rather than as a system whose inner workings could be deduced from its communicative function, or any function. Years later, he would suggest in conversation that phonological systems should be thought of as forms of speech disguise.Footnote ²³ That would explain why rules exist in the first place.Footnote ²⁴ This idea was never written up. It’s possible that Morris came to see the difference between speech disguise techniques, which introduce deliberate obstacles to lexical access, and the typically less disruptive changes caused by phonological processes. In any case, the intuition of phonology operating in part as a form of speech disguise was there at some point.

A linguist who thinks that the whole point of phonological changes, if any, is to actively conceal the input might have difficulty sympathizing with optimality theory (OT; Prince & Smolensky Reference Prince and Smolensky1993). OT grammars assign penalties to deviations from the input. They seek to justify attested changes of the input as the price to pay for improvements in the surface structures. Morris’s second point of divergence with OT stemmed from his belief that the new theory’s chief objective was to eliminate intermediate representations. He and Sylvain Bromberger had argued just a few years earlier for the necessity of multi-step mappings from the lexical entry to the articulatory output.Footnote ²⁵ He considered that issue closed and a theory disqualified if it couldn’t model all patterns describable with rule ordering.

Whatever the source of his reservations about OT, Morris took the work seriously, taught it in his classes, and devoted more time responding to it than he had spent forty years earlier refuting the neo-Bloomfieldian separation of levels. He argued that sound patterns presented as favoring a constraint-based analysis had been misinterpretedFootnote ²⁶ and that OT analyses contained errors inevitably imposed by the wrong choice of theory.Footnote ²⁷ He enumerated empirical results that seemed out of OT’s reach, invoking elsewhere condition effects among them.Footnote ²⁸ He did try to understand the attractions of the constraint-based analysis of reduplication. He failed at that, perhaps because he saw as the only role for correspondence theory that of concealing the inability of surface-oriented markedness constraints to describe sound patterns. (Yet it was Halle and Keyser who had advocated thirty years earlier a constraint-based analysis of meter that relies on proto-correspondence constraints, conditions ensuring a minimal match between the accentual relations present in the text and the rhythm of the abstract verse template.)

While trying to understand reduplication in OT, Morris started looking for alternatives. He worked out an analysis of reduplicant shape and placement that relied not on constraint interaction but on a formalism that would unify metathesis with reduplication. That work promised to deliver new results: a first account of the notoriously hard to analyze Attic reduplication, in Halle Reference Halle, Freidin, Otero and Zubizarreta2008,Footnote ²⁹ and the discovery of a morphological system in which metathesis and reduplication appear to be in free variation, in Harris & Halle Reference Harris and Halle2005.

Morris retired from teaching in 1997. By then he was MIT Institute Professor and former Killian lecturer, the highest honors that his university confers on its faculty; former President of the LSA; the recipient of numerous honorary degrees. He remained active, taught at the 2005 LSA Institute, published books and articles, continued to advise dissertations. He was no longer prowling the corridors in search of students to educate, but former students, including heretical followers of dubious doctrines, would be warmly welcomed into his office. He was eager to understand their work, and would follow without fail if a derivational translation was presented. He continued to fill with severe comments the margins of their manuscripts. He cheerfully faced their objections to his own views. He thought the achievements in SPE had been forgotten, and would rejoice to hear otherwise. On saying good-bye to two former students, on March 28 2018, he remembered, as if to summarize his work, that ‘we did some good teaching’. He died on April 2 2018, surrounded by his family.

In describing the communities from which his own family had come, Morris wrote as follows about the role of the rabbi:

As this description proceeds, it reaches a level of detail that suggests Morris was writing about a particular man, one known to him. He was to us.