Frames, slots, fillers

Levelt (1989), Chapter 9

Utku Turk

LING 449U, How words are built

2026-09-10

Where we left off

• Griffin and Ferreira barely covered morphologically complex words.
• The one place inflection appeared was Property 9, filed next to determiners.
• Levelt Chapter 9 will give us some of the machinery.

Two senses of “morphology”

Levelt’s sense

The affix -ed. Its shape, its syllable, its allomorph.

Our sense

The feature PAST. Case, number, person, honorification.

It is not the case that Levelt does not discuss “our” morphology. He himself insists on the split: distinguish diacritical features such as “past”, “plural”, “third person” from the affixes they induce.

Some examples:

I sould be sheeing him soon

Some examples:

I sould be sheeing him soon [I should be seeing him soon]

The /s/ of seeing was taken early. But seeing did not come out as eeing.

The position and its filler are separate objects.

The unit that moved

I sould be sheeing him soon

shouldʃʊd

seeingsiːɪŋ

• Not two copies of the same /s/ trading places. What exchanged is /ʃ/ and /s/.
• Not shou with see. The rimes stayed where they were.
• The onset is the slot, and only the onset was refilled.

Three levels, three kinds of filler

Level	Frame slots	Fillers
Morphological / metrical	lemma, number, person, tense	lemmas, diacritic features
Segmental	stem, affix	morphemes
Phonetic	onset, nucleus, coda	segments and clusters

The morphosyntactic address

lemmaknow

numberany

personany

tensepast

The morphosyntactic address

lemmaknow

numberany

personany

tensepast

unlocks ~KNEW~

Fill every slot and the lock opens. The output is a morpheme string plus a metrical shape, not sounds.

• One entry eat, several items: eat, eats, ate.
• The diacritics select the item.
• Frequent forms are stored, not composed. Implication? arrives costs no more than arrive.

I'd hear one if I knew it

lemmahear

……

tensepresent

unlocks hear

lemmaknow

……

tensepast

unlocks knew

One possible explanation: The lemmas moved between frames and the tense features stayed. know landed in the past slot and came out as knew, with the irregular form.

Another possibility: “Tense” is dictated by something else, and not lemmas.

raining on a hitchy day

Intended: hitching on a rainy day.

• Realize: rain is a noun, hitch is a verb.

• Can we try to write out the process by which this error happens?
• Can it be the same as the hear vs know?

raining on a hitchy day

Intended: hitching on a rainy day.

• rain is a noun, hitch is a verb. Exchanging lemmas are almost always of the same category.
• So this cannot be the same error as hear and knew.

One level down: stem and root swapped at segmental spellout, where category is no longer visible.

they needed to be maded

1. Level 1 worked. make plus PAST unlocked the stem made.
2. Level 2 failed. The -ed of needed was still available and filled the affix slot.

Level 1 delivered the right stem and the affix still went wrong, so PAST and -ed are two different objects.

Where each error happened

Error	Level	What moved
I’d hear one if I knew it	morphological	lemma, diacritic stayed
raining on a hitchy day	segmental	stem and root
they needed to be maded	segmental	affix only

Surface similarity is no guide to which address failed.

What can move at all

Level

Unit

Example

Moves?

Phonological

distinctive feature

mity the due teacher

no

Phonological

segment

peel like flaying

yes

Phonological

onset, rime, nucleus, coda

face spood

yes

Phonological

syllable

cassy put

no

Morphological

stem, root

raining on a hitchy day

yes

Morphological

affix

they needed to be maded

yes

Morphosyntactic

lemma

I’d hear one if I knew it

yes

Morphosyntactic

diacritic feature

?

?

What do we care about? Morphosyntax or diacritics.

Q: Is there a diacritic exchange or substitution?

• What would an exchange of two diacritic features look like?
• Two features swapping while the words stay in place.

One candidate (questionable)

The key to the cabinets are rusty.

Wrong filler, right slot: PLURAL from cabinets got into the number slot of the verb’s address.

That is a diacritic substitution, not an exchange. The displaced singular does not appear anywhere else.

• Think about agreement again: how does a feature get from cabinets to a verb that far away?
• Are there phonological errors that depend on something equally far away? Vowel harmony?

One more experiment

• Okay, there was one cool experiment in the chapter that I want you to think about.

Treiman’s game

First you learn the items by repeating them, what is the rule you infer?

kig → kaz ig

buf → baz uf

tep → taz ep

nol → naz ol

Take the onset, add az, then say the rime.

skef

Two answers fit the training data.

skaz ef onset cluster held

saz kef cluster broken

skef

Two answers fit the training data.

skaz ef onset cluster held

saz kef cluster broken

Almost all responses were the first. The onset cluster moves as a unit.

In-class exercise

Design one of these, then tell us what it predicts.

1. Run Treiman’s game with L2 speakers whose first language does not allow /sk/ onsets. What do you expect them to answer for skef, and what would either answer tell you?
2. Build a game whose rule is over a morphosyntactic slot instead of a syllable constituent. What is the frame, what is the slot, what is the filler?

Takeaways

1. A frame is an address. Fill every slot and a stored form is unlocked.
2. Level 1 fillers are lemmas plus diacritic features. That is the morphosyntax.
3. PAST and -ed fail independently: knew proves one, maded proves the other.
4. Only units with slots can slip, which excludes syllables and features.
5. The diacritic row of the ladder is still empty. Whether it can be filled is an open question, and a testable one.

Bring questions

• Where do you put case, honorification, or evidentiality in this diagram?
• Levelt’s model stores every frequent inflected form whole. What does that leave for morphology to do?
• If agreement attraction is a filler error, what predicts which noun controls the verb?

Extra slides

The word sketch

TOT states: you have the metrical shape and the first sound, not the word.

• First phoneme correct 60 to 70 percent of the time.
• Syllable count correct 60 to 80 percent, plus which one is stressed.
• A phonological blocker induces TOT. A semantic one does not.

Form comes out in layers, and the layers can come apart.

The Unit-Similarity Constraint

The intruder is of the same level and the same category as the target.

• Lemmas with lemmas, stems with stems, affixes with affixes, onsets with onsets.
• Stems never exchange with affixes.
• Affixes never exchange with onsets.