How many animals do you see?

the "two dogs" example

This is intended to illustrate how the lexicon is accessed, words formed and pronunciation movements "computed" -- as information from the mind, lexicon, morphological and phonological rules smoothly work together to get the appropriate movement. This is one place where "co-articulation" effects are determined

I have left out the syntactic processes that would embed the "dogs" into a phrase "two dogs" or a sentence "I see two dogs".

Terms and concepts

the "mental dictionary" (MD)

It represents what we know about individual morphemes (words). Desk dictionaries are modelled after the MD! What is in them?


graphic representations of speech

time-pressure wave

A simple plot of how pressure outside the mouth or wherever the microphone is varies during speech. (dB x time)

See the plot above (blue ink) of "dogs"; from the axis we see it took me about 400 milliseconds to utter it.


intensity (dB indicated by darkness at x-y coordinates) at various frequencies (Hz on y-axis) over time (milliseconds on x-axis)

the actual plot requires a mathematical analysis of the time-pressure wave to determine its frequency components. A method called Fourier analysis is used.

In the above example spectrogram of "dogs", there are at least four visually distinctive segments -- an initial segment corresponding to the release and "voice onset" of [d], a longer segment of distinct frequency bands in dark ink showing the significant energy (formant structure) of the vowel [o], and the final stop [g] with no energy present, then the distinctive high frequency noise around 4000Hz. corresponding to the inflection [s].

Fourier analysis

Jean Baptiste Fourier (1768-1830) figured out how to decompose certain complex waveforms into a set of weighted sine waves that when added together reconstructed the complex waveform. This is essentially what a spectrogram displays.

Chorus metaphor for the voice

(speech is produced with various frequency components -- bass, tenor, etc. soprano -- as if by a chorus all signing the same thing.)

two - tone speech vs normal speech

(Demo from Haskins Lab of Pinker's example, ch. 6. Speech synthesis by just two varying sine waves)

the frequency range of normal speech

100 to 5000 Hz (See spectrogram above). One can think of speech being simulated by lots of tones being turned on and off according to a musical score -- the chorus metaphor.

the vocal tract

consonants described in terms of manner and place of articulation

(a complex description in terms of the IPA symbols) See their web site and chart of symbols.

the larynx

evolutionary history -


what information its movements convey

phonological rules

These rules modify distinctive features of the phonological segments according to context. In the above example they tend to voice fricatives that follow voiced consonants like /g/ and others /b/ and /d/. A voiced consonant is one that is immedately followed by phonation -- vibration of the larynx. An unvoiced consonant is one that the voicing of the following vowel is delayed maybe 100 milliseconds -- /p,t.k/ that correspond to the voiced /b,d,g/.

The rule that is illustrated above also operates on formation of English tenses and possessives since both of these have an inflection /-s/ added to stems that may have final voiced or unvoiced consonants. Compare "the fishes", "She fishes" and "the fishes' tail".

As Pinker says.179, "..phonological rules see features, not phonemes, and they adjust phonemes not features... A phoneme is a bundle of features."