Editor : D. BouwhuisAdviser: H.J.A. van BeckumTypist: Miss J. Swinkels-Kool

The picture shows a tight spot in the word PROGRESS; it in­

dicates one of the positions which the eye is fixating dur­

ing reading. A study on eye movements during reading appears

on page 30.

INSTITUTE FOR PERCEPTION RESEARCH - INSTITUUT VOOR PERCEPTIE ONDERZOEK

INSULINDELAAN 2 EINDHOVEN HOLLAND TELEPHONE 040-756605 / 040-472485

ORGANISATION

supervisoryboard i.p.o.

Ir. K. Kooij (chairman)Drs. J.H. BannierProf. Dr. J.F. SchoutenDr. Ir. K. TeerProf. Dr. W.A.T. Meuwese

scientific board i.p.o.

Prof. Dr. H.B.G. Casimir (chairman)Prof. Dr. M.C. ColenbranderProf. Dr. J. Droogleever FortuynProf. Dr. N.H. FrijdaProf. Dr. L.B.W. JongkeesProf. Dr. J. KoekebakkerProf. Dr. S.L. KweeProf. Dr. Ir. H. MolProf. Dr. A.J.B.N. ReichlingProf. Dr. R.J. RitsmaProf. Dr. P.C. VeenstraProf. Dr. A.J.H. VendrikProf. Dr. Ir. C.J.D.M. VerhagenDr. Ir. P.L. WalravenProf. Dr. P.J. WillemsProf. Dr. Ir. A. van Wijngaarden

director i.p.o.

Prof. Dr. C.A.A.J. Greebe

research associates

D.J.H. AdmiraalJ.J. AndriessenDr. H. BoumaDrs. D.G. BouwhuisDrs. B.L. CardozoDr. H. DuifhuisDr. J.P.M. EggermontIr. F.L. EngelJ. I t HartDrs. A.F.V. van KatwijkIr. W.G. KosterF.F. LeopoldG.J.J. MoonenH.F. MullerDr. Ir. F.L. van NesDr. S.G. NooteboomDr. Ir. J.A.J. RoufsI.H. SlisJ.C. ValbrachtIr. L.L.M. VogtenJ. VredenbregtIr. L.F. Willems

II

IPO annual progress report 8 1973

- Eindhoven- 's-Gravenhage- Eindhoven- Eindhoven- Eindhoven

- Heeze- Utrecht- Groningen- Amsterdam- Amsterdam- Baambrugge- Eindhoven- Amsterdam- Amsterdam- Groningen- Eindhoven- Nijmegen- Delft- Soesterberg- Tilburg- Amsterdam

research staff

M.A. AlewijnseH.J. BleilevenP.B.W. BoonstraTh. M. BosTh.A. de JongC.A. LammersG.H. van LeeuwenA.C. van NesW.H. NoordermeerJ.A. Pellegrino van StuyvenbergA.L.M. van RensR. van SchuurH. de Vries

secretaries

Mrs. L.J. Savenije-ClignettMiss C.J. SenkeldamMrs. J.W. Tielemans

library

Mrs. J.M. HoogervorstMiss C.W. KoningMrs. J.P.M. Swinkels-Kool

workshop

M.A. van den BanC.G. BastenJ.H. BolkesteinP.A.N. BroekmansC.Th.P. GodschalxG.C. JonathanH.E.M. MelotteD.J. van der Wees

advisers

Prof. Dr. A. Cohen (University of Utrecht)Dr. G.J. Fortuin

guest researchers

Miss C.M.J. Hulst (part-time) Occupational Medical Dept. of theNederlandse Spoorwegen (Netherlands Railways). Utrecht

Ir. L.P.A.S. van Noorden, Netherlands Organisation for theAdvancement of Pure Research

Drs. C.W.J. Schiepers, Netherlands Organisation for the Ad­vancement of Pure Research

student assistent

P.J.J. Hamilton

III

IPO annual progress report 8 /973

This report or any part thereof may not be reproduaed in any form without the

written permission of the Institute for Peraeption Researah. Reprints of the

separate contributions are available. Illustrations may be reproduced only

with explicit mentioning of source; copies will be appreciated.

IV

IPO annual progress report 8 1973

INTRODUCTIONThe present Chairman of the Supervisory Board, Mr. K. Kooij, expressed the gratitude

of all present to its retired members: Professor Dr. H.B.G. Casimir, Dr. G.J. Fortuin

and Professor Dr. C. Zwikker on the occasion of a joint meeting of the SupervisoryBoard and the Scientific Board held in May 1973.

Their many contributions to the well-being of the Institute have extended over long

periods of time. Professor Zwikker and Professor Casimir have even been Chairman and

Member respectively from the foundation of the Institute for Perception Research in1957.

Who could have thought on that occasion that we were to lose, so soon afterwards, the

cooperation and friendship of Dr. Fortuin who remained an adviser to the I.P.O.? Tothe distress of all of us he died in January, 1974.It will be concluded from the significant decrease in size which this issue shows in

comparison with its predecessor that our objectives in publishing the Annual ProgressReport have shifted. We have not tried in any way to strive after a complete and self­contained account of our activities.

In the introduction to the previous Progress Report I already alluded to the view

that one of our purposes in publishing this Report should be to "provide the reader

with a context in which publications from the Institute can be placed".Pursuing this train of thought we again present some papers which are primarily to be

looked at as reviews of the work which has been done in the I.P.O. Furthermore, ab­

stracts have been included in our list of publications. In this way we have tried toexpress our readiness to provide the reader with reprints of any I.P.O. Publicationsin which he may take an interest.

C.A.A.J. Greebe

v

IPO annual progress report 8 1973

CONTENTS

organisation

introduction

contents

1 auditory perception

B.L. Cardozo

I.P.O. Auditory Research in 1973

2 speech

S.G. Nooteboom, I.H. Slis and L.F. Willems

Speech Synthesis by Rule; Why, What and How?

3 visual perception

H. BoumaResearch on Vision

C.W.J. Schiepers

Length Estimation of Letter Strings

J.J. Andriessen and A.H. de Voogd

Analysis of Eye Movement Patterns in Silent Reading

F.L. Engel and T.M. Bos

Visual Conspicuity Related to Generalisation of Background

Regularities

4 perceptual and motor skills

W.G. Koster

5 instrumentation

D.J.H. Admiraal

Introduction I.P.O. Instrumentation 1973

D.J.H. Admiraal

CIDAT, A Continuously Adjustable Timer with Digital Display

G.J.J. Moonen and Th.A. de Jong

MARIE; Interface between Computer and Experiment

6 i.p.o. publications 1973

VI

IPO annual progress report 8 1973

page

II

V

VI

3

14

25

30

36

43

50

52

55

58

1 auditory perception

I.P.O. AUDITORY RESEARCH IN 1973B.L. Cardozo

This report on progress in auditory research at I.P.O. in 1973 ~ill be limited to

this brief survey; work has continued mainly along existing lines of investigation.

The perception of tone sequences was studied by Van Noorden in several ways. His

experiments can be divided roughly into two groups. The first deals with the percep­

tion of time patterns and the second with the perception of coherence of the elements

of a tone sequence, i.e. fusion.

As regards the perception of the time pattern, it was known already that in a long

sequence ABA ABA ... the accuracy with which a tone can be adjusted to its central

position along the time axis depends upon the frequency jumplfB/fA - 11 (Van Noorden,

1971 a). However, when the sequence was reduced to one single triplet ABA, then the

timing error was found to be much smaller under comparable conditions, and to be less

dependent upon the frequency jump. Extending the stimulus to AABAA, to AAABAAA, etc.

was found to bring the timing error back to its original order of magnitude and to

restore the dependence upon the frequency jump. This "embedding effect" can hardly

be due to a peripheral factor.

Coherence in tone sequences was studied under a wider range of tone rates than used

previously. To a large extent, the experiments were performed on sequences of the

type ABABAB .. , A and B being pure tones with a duration of the order of 50 ms. The

inner and outer fusion boundaries were measured to depend upon tone rate in a way

that is qualitatively similar to the result obtained earlier with ABA ABA ... se­

quences (Van Noorden, 1971 b). However, when making the sequence slower than 2 tones

per second one finds that the inner fusion boundary shifts towards larger frequencyjumps,

The study of coherence was further extended to alternations BCBC .. of a (harmonic)

complex tone C with a pure tone B and also to alternations of tones of unequal am­

plitudes. Preliminary results on the BCBC stimulus indicate that coherence between a

residue and a pure tone of the same pitch can occur only in relatively slow tone

sequences. In the stimulus with alternating amplitudes fission was found in sequences

with a loudness difference 6 L of 3 dB and more even with monotone sequences. An

existence diagram of coherence in ABAB " sequences was determined in the 6 L - 6 f

plane with tone repetition time TO as"a parameter. In very fast alternations BCBC ",

fusion between B and a single component of C can be heard when the amplitude of B is

adjusted. A full publication of the perception of tone sequences is being prepared

by Van Noorden.

Pure tone masking experiments were continued by Vogten, who uses a phase-locked tone

burst as a probe and a continuous pure tone as a masker. With fixed probe frequencies

f p = 0.5, 1, 2, 4, and 8 kHz, "masking curves" were measured either as iSO-Lp curves,

showing the masker level Lm as a function of the masker frequency f m for fixed probe

level L ; or as iso-L curves, showing L as a function of f , with fixed maskingp m p m

level Lm, Previous findings (Vogten, 1972) were confirmed and extended for other sub-

jects. Iso-L curves for L = 10 dB 5L show an interesting similarity to physiologi-p pcal tuning curves. Again asymmetries in both iso Lp - and iso Lm-curves were found at

all frequencies studied. The words iso-L curve and iso-L curve have been intro-m pduced in order to avoid confusion with the conventional masking curve, that refers to

measurements in which the probe frequency is variable and the masker frequency is

,PO annual progress report 8 /973

2

constant. In terms of the conventional masking curve, the asymmetries amount to the

following: at low sound levels, the maximum masking effect is not found at the masker

frequency f m but at a frequency that is about 8\ lower than f m; at high sound levels,

the maximum masking effect is found 10 to 20% higher than f m. It remains an open

question whether this behaviour can be attributed to properties of the excitation

function at the level of the auditory receptor.During his stay at M.I.T. Duifhuis (1973) has worked out a quantitative theory of

backward masking. The theory is based on a logical extension of earlier modellingwork of the peripheral auditory system (Duifhuis, 1972), and applies detection theory

to the level of neural discharge information as proposed by Siebert (e.g. 1965).

references

Duifhuis, H. (1972) Perceptual Analysis of Sound, Doctoral Thesis, University ofTechnology, Eindhoven

Duifhuis, H. (1973) Consequences of Peripheral Frequency Selectivety for Non­Simultaneous Masking, to be published in J. Acoust. Soc. Amer.

Van Noorden (1971 a) Discrimination of Time Intervals Bounded by Tones of DifferentFrequencies, I.P.O. Annual Progress Report, ~, 12 - 15.

Van Noorden (1971 b)Rhythmic Fission as a Function of Tone Rate, I.P.O. AnnualProgress Report, ~, 9 - 12.

Siebert, W.H. (1965) Some Implications of the Stochastic Behavior of PrimaryAuditory Neurons, Kybernetik, ~, 206 - 215.

Vogten, L.L.M. (1972) Pure-Tone Masking of a Phase Locked Tone Burst, I.P.O. AnnualProgress Report, ~, 5 - 16.

3

SPEECH SYNTHESIS BY RULE; WHY, WHAT AND HOW?S.G. Nooteboom, I.H. Slis and L.F. Willems

There are more than one reasons why speech researchers may be interested in systems 0

speech synthesis by rule. Such systems eventually achieve practical importance to

voice response units in man-computer communication, as part of a reading machine for

the blind, a spoken encyclopaedia or to the automatic generation of taped sets of

spoken instructions for certain tasks such as wiring telephone exchange units (e.g.

Flanagan, Coker, Rabiner, Schaefer, Umeda, 1970).

One may also be interested in speech synthesis by rule as a research tool in the

domain of phonology and speech perception. The importance of synthesis by rule in

this respect has been stressed a number of times (by, among others, Liberman,

Ingemann, Lisker, Delattre and Cooper, 1959, Lisker, Cooper and Liberman, 1962,

Mattingly, 1971, Klatt, 1971, Holmes, 1972). It is this interest which constitutes

the main motivation for the work in our Institute on a system of speech synthesis

by rule. As this work involves considerable effort it seems worth while to give some

thought to such questions as: Why is synthesis by rule a desirable research tool?

What properties should the system have? How can these be achieved? In an attempt to

answer such questions let us restrict our considerations to terminal analog synthesis.

Operationally, we define a system of speech synthesis by rule as a system which

accepts a discrete, typed input in terms of phoneme-like symbols and some additional

symbols (e.g. stress marks, word, morpheme and phrase boundaries) and automatically

converts this input into intelligible speech.

why?

We will put forward three condiserations which, in our view, make it desirable to work

on a system of speech synthesis by rule.

a. Generation of stimuli

Such a system may provide the means of rapid and, easy generation of large sets of

stimuli for perceptual experiments. In principle, all stimuli which can be made by

a rule system, can also be made by ad hoc specification of the parameters. It is

obvious, however, that, if we wish to generate large sets of stimuli, especially

if the stimuli consist of whole words, word groups or sentences, this soon

becomes very laborious. Once a suitable rule system, or even part of it, has been

developed, many experiments can be carried out in much less time than before.

b. Heuristic strategy

The second consideration in favour of speech synthesis by rule as a research tool

might well be the most important. Working on synthesis by rule is a powerful and

inspiring heuristic strategy. In attempting to synthesise intelligible and

reasonable sounding speech we are forced to make explicit hypotheses concerning

perceptually relevant acoustic properties of speech. In the frequent failures of

our attempts we are immediately confronted with many things we do not know about

speech. In this way we readily run into research problems we would not have

thought of otherwise, or would not have considered seriously, but which, never­

theless, may be of fundamental importance to understanding the processes of

speech production and perception.

An example is provided by the growing amount of experimental work on intonation

4

and temporal organisation of speech in a number of speech laboratories. The re­

newed interest in this domain seems at least partly inspired by the failure of

existing synthesis by rule systems to shape the pitch contours and temporal struc­

tures of speech in a perceptually satisfactory way. This leads not only to a

search for useful prosodic rules but also to the more important question of the

perceptual part played by prosodic patterns in speech.

Another, possibly related, research problem which comes into focus in work on

speech synthesis by rule is the problem of what we would like to call "perceptual

integration", All synthesis systems known to us exhibit in their output signals

instances of sound segments which perceptually do not seem to be part of the

perceived speech patterns, and which are difficult to locate in the utterance.

We may say that they cannot be "perceptually integrated" with the rest of the

speech into recognisable patterns. This has a disturbing effect on speech intel­

ligibility. It also immediately raises the question: What perceptual mechanisms

are responsible for our ability to hear normal speech as perceptually integrated

patterns, and under what conditions do these mechanisms fail?

A third and final example is of a more linguistic nature and concerns the input

of the rule system. An unstructured string of input symbols corresponding to

phonemes or speech sounds is not suitable for being converted into intelligible

and acceptable connected speech. Relevant information concerning morpheme, word

and phrase boundaries, lexical stresses and pitch accents would be lacking. This

information is needed for the shaping of perceptually satisfactory pitch contours,

temporal structures, and for inserting speech pauses in the correct places. The

question we are confronted with here is what information should be provided at

the input, and what information can be found by automatic analysis of the input

string. If the system is supplied with word boundaries, in many cases lexical

stresses, morpheme boundaries and phrase boundaries may be found by automatic

analysis, but as yet no way has been found to predict the correct positions of

pitch accents. It seems that information on both the syntactic and the semantic

structure of the phrase or sentence is needed in order to do so. The basic rules

involved still escape explicit formulation, however.

e. Making and testing of complex models

The relation between a linguistic, discrete representation of the sound level of

language, in terms of phonemes or speech sounds, and an acoustic specification

of speech is a complex one. Speech synthesis by rule provides an excellent means

of making explicit models of this relation, and of keeping the models testable.

The acoustic regularities of speech are the result of interactions of a great

many factors. In accounting for these interactions we need intricate rule systems

that can rapidly be revised and tested both quantitatively and perceptually. One

of many possible examples is provided by the temporal organisation of speech.

Actual segment durations in speech result from interactions between the feature

composition of the segments, the internal structure of the syllable (for con­

sonants especially the structure of the consonant cluster), degree of stress,

position in word and phrase and other factors. Such interactions can be modelled

in a set of quantitative rules, but without implementing these rules in a syn­

thesis system it would be difficult to test them as to their acoustic and per­

ceptual effects.

5

Summarising, we may state that it is desirable to work on speech synthesis by

rule (a) in order to obtain a rapid and easy way of generating acoustic stimuli

for experiments on speech perception, (b) as a heuristic strategy in speech

research, (c) for modelling intricate interactions underlying acoustic regulari­

ties of speech, and testing these models both acoustically and perceptually.

what?

Below we formulate some requirements and properties desirable for a system set up

for the purposes outlined above.

a. Ease of operation

A synthesis system as we conceive it is used in the laboratory by more people

than only those actually involved in its development. All researchers of the

laboratory staff, guest workers and students who wish to use the system must be

in a position to do so. This implies that operation should be easy and easily

acquired.

b. Visual, aaoustia and numeriaal feedbaak

For all purposes mentioned it is important not only to have the facility to listen

to the output and tape it, but also to have an immediate check on the parameter

values generated, both visually on an oscilloscope, and numerically in a printed

output. One should be able to select for feedback the parameter values one is

interested in.

a. Speaial input for parameter values

It is desirable to have facilities for controlling the value of one or more

parameters on the spot without changing the rule system. This may concern an

acoustic parameter for one particular segment in the speech chain, the rest being

controlled by the rule system, or a rule coefficient of the system, affecting all

segments to which the rule applies. The desired parameter values could be supplied

to the system either by a typewritten input or by some other manual control.

d. Exahangeability of rules and subsets of rules

It is desirable to be able to suppress each rule or subset of rules temporarily

and replace it by a new one. This holds for all levels of organisation, which

implies that the system should have a clear organisation in smaller and larger

subsets of rules, all with their own names, and as independent of the other subsets

of rules as possible. This would enable us to compare and test alternative rules

or rule sets.

We may wish, for example, to replace the rules for vowel durations in stressedsyllables, or the whole set of rules governing segment durations, or all prosodicrules.

e. Retraaeability of aaoustia effeats

The system should be so organised that it is, in principle, always possible to

retrace acoustic phenomena in the output to the rules or rule interactions

responsible for them. This requirement is extremely important and extremely dif­

ficult to meet. It is important for a rather self-evident practical reason. If

it is not met, it will often be very difficult to correct defects in the systembecause one simply does not know which rules to change. It is also important for

6

a more theoretical reason, because if it is not met, we cannot consider the rule

system to be a useful, insight giving model of the interactions underlying acous­

tic regularities in speech. We cannot use the rule system to explain how these

interactions work. The rule system itself becomes yet another complex phenomenon

to be explained.

This danger is real. A more or less complete set of rules for the generation of

speech soon becomes so complex that even the people actually building the system

may easily lose track of the chains of causes and effects in the system. This

should be avoided as best as possible by a rigid and clear organisation. The

inclusion of ad hoc rules, the interaction of which with other rules is not known,

should be considered bad practice. In this respect it is important to keep up an

accurate and easily readable description of the system and its workings.

f. Compatibi~ity with speech production mode~s

Although we are limiting our considerations to terminal analog synthesis, it is

good to keep in mind that many of the regularities the system has to simulate in

real speech stem from properties of the human speech production system. The

explanatory power of our rule system will be considerably greater if it is so

organised that specific rules or subsets of rules can be related to existing

models of speech production.

For example, in writing rules for formant movements, it seems advisable to in­

corporate constraints which follow from dynamic models of speech production. It

might even be worth while to generate separately the effects of the dynamics

of articulatory movement and those of the acoustic theory of speech production

in order to enhance the generalising power of the rules.

Similarly, it seems worth while to write separate rules for prosodically condi­

tioned durational variation and durational variation caused by the short term

dynamic behaviour of the articulatory organs, and make lower level interaction

rules specify the actual durations.

In writing rules for consonant clusters we may introduce units corresponding to

articulatory segments, which, owing to the considerable overlap in time of

articulatory gestures, are not directly reflected in the acoustic signal, but

which help in writing rules with some generalising power.

Keeping our rules as much as possible compatible with speech production

models may guide us in creating a rule system which has a clear internal organi­

sation and which is an explicit formulation of non-trivial aspects of what we

know about speech.

~w?

In this final section of our paper we describe briefly how our system for speech

synthesis by rule is organised and how we have implemented some of the desired pro­

perties described above. This system replaces an older one which was based on a

rather extensive hardware machine, the IPOVOX II, to which a software rule system

was added. This older system has been described earlier (Slis and Muller, 1971,

Slis 1971). In the present system the hardware has been confined to the actual

signal generator, a digital hardware synthesiser (Rockland), which is driven froma computer (Philips P 9202, 16 bits, 16 K). A disk memory is used for additional

storage capacity. The Rockland synthesiser needs fresh information about all para-

7

meter values at each period of the fundamental frequency ("glottal period"). Thus

in the rule system the input in terms of discrete phoneme-like symbols has to be

transformed into a sequence of segments corresponding to glottal periods, each withappropriate parameter values. Our rule system is at present conf~ned to Dutch.

a. Segmental organisation of the programme

A main feature of the programme is that the segmental organisation of the input

is sustained as long as possible in the rule system. Thus most of the synthesisrules operate on segments of phoneme size, each such segment in principle corres­

ponding to an input symbol. For practical reasons this correspondence is not

quite kept up in the case of vowels. All vowels consist of two symbols in theinput. The rule system, however, interprets a phonologically short vowel as one

phoneme-like segment, and a phonologically long vowel or diphthong as two seg­ments.

b. Changing parameter values

For each phoneme of Dutch, standard values for a set of 20 parameters are stored

in a table. These values do not necessarily correspond to any realisation ofthe phoneme but are rather so chosen that the rule system is optimised. Theparameters are: Vowel formant frequencies F1 , F2 , F3 , F4 , FS and their band widths

B1 , B2 , B3 , B4 , BS' a nasal formant FNI and a noise formant Fns with their res­pective band widths BNI and Bns ' amplitude of voice, noise and hiss (after

formant filtering) AV' ANs , AAsp' pitch level FO and segment duration DUR. Thetwentieth parameter is not used at present.

Each parameter is assigned

three values, viz.:1. A target value (TVALUE in

Fig. 1).

2. A time constant (TeON inFig. 1), controlling theduration of the transition

from a previous parametervalue (O-VALUE in Fig. 1)

to the present target

value.3. A value (TIM in Fig. 1),

controlling the moment of

onset of the transition

t=duration

T-value

-III

I,I

I 1TIM~TCO~

I :

- - - - - - - - duration - - - -

representation of the time coursethe synthesis parameters.the previous target valuethe target value of the presentsegmentthe onset time of the change to­wards the new target

= the duration of the change towardsthe new target

t=O

TeON

TIM

Schematicof one ofO-value =T-value =

O-value

from one parameter valueto the next, with respectto an abstract phoneme

boundary (t = 0 in Fig. 1). If the transition starts before t = ~ TIM is assigned

a negative value. In Fig. 2 an example is provided of some parameter values andtheir transitions in a plosive-vowel combination. This may give an impression of

the possibilities of the system. The transitions may also follow more complex

functions than provided by straight lines.

8

FNs

TPDUR

TPAAs =0

TPF1 =400 Hz

TPANs=O

TPF2 = 800 Hz

TPF3 =1050 Hz

TPAV= 1000

T1.F1

AV

AAs

ANs ~

F1

F3

FO

F2

-20 -10 0 10 20 30 40 50 60 70 80 90 100 110 T-+

Parameters TP=Target TC=Dur. of TI=Start of(TVALUE) change (TCON) change (TIM)

FNs =noise formant TP.FNs TC.FNs TI.FNsF3 =third formant TP.F3 TC.F3 TI.F3F2 =second formant TP.F2 TC.F2 TI.F2F1 =first formant TP.F1 TC.F1 TI.F1AV =amplitude voice TP.AV TC.AV TI.AVFO =fundamental freq. TP.FO TC.FO TI.FOAAsp=amplitude aspi-

ration TP.AAsp TC.AAsp TI.AAspANs =amplitude noise TP.ANs TC.ANs TI.ANs

Fig. 2. Example of the-time courses of 8 of the synthesis parameters in a plosivevowel combination.

9

o. Btook diagram of the system

Fig. 3 presents a block diagram of the system. The system is so organised that

there are four separate programmes and three data storage tables on the disc.

The division into separate programmes is as follows:

1. INPUT part. The input string is stored and the conditions, calculated from the

input string, are added to each phoneme.

2. RULE part. Standard parameter values are supplied from storage table. Input

conditions of f-1, f, f+1, f+2 are made available. Synthesis rules operate on

standard parameter values of segment f.

3. SEGMENT TRANSFORMATION part. The data for each phoneme, calculated in the RULE

part and stored, are reorganised in terms of data for glottal period segments.

4. EXECUTION part. The glottal period data, calculated by the SEGMENT TRANSFOR­

MATION part and stored, are transmitted to the speech synthesiser.

PROGRAMMES DATA STORAGE ON DISK

INPUT PART: TABLES WITH STANDARD PHONEMES:

conditions are added to target values onsets andthe phonemes of the input of parameters durations ofstring F1 B1 transitions.. F2 B2 to new targets

RULE PART: F3 B3 AV FO

.... F4 B4 ANs AAspthe phonemes conditions of .... F5 B5 DURf-1, f, f+1, f+2 are avail- FNl BNl FNs BNsable; the parameter valuesfor the segment "f" arecalculated in a set of ... PHONEME SEGMENTS:subroutines ...

... table with targetand time values for

SEGMENT TRANSFORMATION PART: ... the synthesis of

the segments of phoneme size.... phoneme segments

are transformed into segmentsof glottal period size ... "GLOTTAL PERIOD" SEGMENTS:.. ...

table with target and

EXECUTION PART: time values for the.... synthesis of glottal

the glottal period segments .... period segmentsare transported to thespeech synthesiser ..IHARDWARE SYNTHESISER I

(ROCKLAND)

Fig. 3. Block diagramme of the synthesis by rule system.

Of these four separate parts of the system the EXECUTION part is trivial. Below

we consider in somewhat more detail the inputs to the RULE part, the internal

organisation of the RULE part, the function of the SEGMENT TRANSFORMATION part,

and some additional features that have to do with the flexibility and ease ofoperation of the system.

10

d. The inputs to the RULE part

The RULE part has two different inputs, one being supplied by the table with

standard phoneme parameters, the other by the INPUT part of the system.

As said above, the table with standard phoneme parameters contains maximally

20 parameters for each phoneme, each parameter being assigned 3 values, viz. target

value, duration of transition, and onset of transition. These values are abstract

with respect to phoneme realisations, and are chosen so as to optimise the rule

system. Perhaps they most closely resemble phoneme realisations in optimally

pronounced monosyllables, but they are not identical.

The memory space for one of the parameter values is actually used for a different

purpose. A 16-bit memory word, belonging to the table with standard parameters,

is filled with information on the feature composition of the phoneme. The features

used are quite straightforward distinctive features, such as place of articulation,

voiced/voiceless, degree of opening, etc. The classification obtained in this way

is extremely helpful in writing efficient, generalising rules. This also helps us

in keeping the rule system compatible with speech production models.

The INPUT part of the system accepts a typewritten input consisting of phonemes

and some additional information coded in normal ASCII characters. All consonants

are presented by one symbol, all vowels by two, viz. the long vowels and diph­

thongs by two letters, the short vowels by a letter plus a point. E.g.

/a:/ = AA, /au/ = AU, /1/ = I.

Additional symbols and their meanings are listed below:

blank word boundary primary lexical stress

morpheme boundary " secondary lexical stress

syllable boundary caesura

+ important word ? end of utterance question

t pitch accent end of utterance

The symbol for 'important word' should precede the word concerned, symbols con­

cerning stress and accent should precede the syllable concerned.

In the INPUT part the information supplied by the additional symbols is added

as conditions to each phoneme concerned. Some further conditions are calculated

from properties of the input string, for example number of syllables in the word,

number of syllables remaining to be produced in word, stress group and phrase,

number of pitch accents remaining to be produced in the phrase, etc. Some con­

ditions having to do with the internal structure of the syllable and the position

of the phoneme in it, are also added.

The INPUT part supplies the RULE part with information in a phoneme-by-phoneme

way. Together with the conditions 6f the phoneme considered it makes available

the conditions of the preceding phoneme and the two following phonemes.

e. The internal organisation of the RULE part

In Fig. 4 we have attempted to give a schematic account of the internal organisa­

tion of the RULE part of the system. The leftmost block is in itself a trivial

part of the system, calling on the necessary information and the subroutine rules.

The whole block called 'subroutine rules' is the interesting part of the system.

It can be replaced by any appropriately formulated alternative set of subroutine

rules. As it is, the subroutine rules are hierarchically organised. An incomplete

impression of this hierarchical organisation is presented in Fig. 5. The sub-

11

TABLES WITH PARAMETERS

RULE PART OF THE SYNTHESIS PROGRAMME • FOR STANDARD PHONEMESON THE DISK

(information on phon.-1, phon., phon.+1,phon.+2, syll.-1, syll., and syll.+1is present)

+SUBROUTINE RULES (exchangeable)

I

<. important word? >... syll. with "- JFO rulesl~pitch acoent ?/ I

IIduration rules~

syll. with ? lexical accentllexical accent rules

Irules for ;\ next syll. with ? duration rulesr-unimportant words pitch accentI

next syll. with ? lexical accent~lexical accent rules

syll. with? no-accent rulesr-no-accent

!place-in-word ruleslI

IFO rules for place i4the utterance

IIduration rules for IIplace in the utterance

I\context rules for Iconsonants and clusters

I\context rules forivowels

ITABLE WITH SYNTHESISPARAMETERS ON THE DISK

Fig. 4. Flow diagramme of the rule part of the synthesis system.

routine rules are organised in a number of separate blocks, named in Fig. 4, such

as FO rules, duration rules, lexical accent rules, consonant and consonant cluster

rules, etc. Each of these blocks can be replaced by an alternative block. Each

block is split up into a number of smaller blocks. This is exemplified in Fig. 5

for the block DURATION RULES for place in the utterance. The hierarchical orga­

nisation of the subroutine rule system helps us in keeping track of the internal

workings of the system and makes it possible to exchange rather easily subblocks

of subroutine rules or individual rules. In sbme cases, however, things may

become somewhat more complex than they appear to be, owing to necessary inter­

actions between blocks of subroutine rules, as, for example between FO rules and

duration rules. It may be necessary to cause FO movement to affect duration, and

12

MAINSYNTHESISPROGRAJvIM:E

EXCHANGEABLESETS OF RULES

SD1lROUTINESFOR DIFFERENTRULES

SD1lROUTINESFOR PARTSOF RULES

SEGMENT EXECUTION -.INPUT part r---. RULE part -+ TRANSFORMATIONpart part

~ ~ or~,

ISD1lROUTINEI SD1lROUTINEI SD1lROUTINEIRULES (1) RULES (2) RULES (3) ..............

'W '-J; + 'tFO RULES DURATION RULES

pitch ace. ........ . for place in .................the utterance

+ + I I I +ADVANCE INSERT INSERT INSERT ASK DECODEonset of caesura amplit. amplit. phoneme syllablesegment segment segment decay label conditions

see:ment

Fig. 5. Illustration of the hierarchical structure of the rules.

vice versa.

f. Opepation of the SEGMENT TRANSFORMATION papt

Our hardware synthesiser has to be supplied with fresh parameter values every

new glottal period. For this reason the phoneme size segments have to be split up

into the appropriate number of smaller segments, corresponding to glottal periods.

Thus the duration of these smaller segments is derived from the FO contour, as it

is determined by the FO rules. During voiceless phoneme segments information on

the fundamental frequency has to continue, as it were in a virtual form. After

a voiceless segment the next segment, if voiced, has to start with a full glottal

period. Abrupt changes in parameter values can only be made at the beginning or

end of a real or virtual glottal segment, not in the middle of a glottal segment.

Thus the temporal accuracy of the system depends on the instantaneous glottal

period length.

g. Some additional featupes

As the system is meant to be used by many persons for various different purposes,

special attention has been given to the design of additional features enhancing

the flexibility and ease of operation.

The operation of any rule or block of rules in the system can be suppressed

manually by means of a set of sense switches. This is helpful both in making sti­

mulus tapes and in playing around with the system for heuristic purposes.

Parameter values and coefficients of selected rules can be manually controlled

with a set of potentiometers (connected to the computer with an ADC). This can be

done at various levels of the programme. For example, one of the parameter valuesin the table of standard phonemes, or the value of a coefficient on one of the

rules can be brought under manual control. In the SEGMENT TRANSFORMATION part of

the system a segment parameter in the already asse~bled utterance can be replaced

in this way. This provision gives a flexible and fast way of interacting with the

13

system. It can also be used in on-line experiments, in which subjects are asked

to adjust the parameter or coefficient value according to some perceptual crite­

rion.

The quantitative effect of any variation in the synthesis fu~ctions can be checked

in an optionally printed output. An immediate visual check of the synthesis func­

tions can be obtained with the aid of a storage oscilloscope attached to the

system.

final remarks

In the sections of this paper called Why? and What? we gave some general considera­

tions concerning systems for speech synthesis by rule. In the section How? we gave

a more or less systematic account of some properties of the system we are actually

working on. This account is far from complete for two reasons. First, the individual

rules of the system are far too many to give an explicit account of them within the

space assigned to us. Secondly, the system is not yet complete itself, in the sense

that it does not yet generate intelligible and acceptable acoustic wave forms for

all possible input strings. The basic organisation of the system as described in

this paper is complete, however, and we would like to emphasise once more that even

with an incomplete set of synthesis rules many interesting experiments can be carried

out. Although the system came in working order only a short while ago, it has al­

ready proved its usefulness as a research tool.

references

Flanagan, J.L., Coker, C.H., Rabiner, L.R., Schaefer, R.W. and Umeda, N. (1970)Synthetic Voices for Computers, IEEE Spectrum, 2, nr. 10, 22 - 45.

Holmes, J.N. (1972) Speech Synthesis, London, Mills and Boon.

Klatt, D.H. (1971) A Theory of Segmental Duration in English, Paper presented at the82nd Meeting of the Acoustical Society of America, Denver, Colorado, October19 - 22, 1971.

Liberman, A.M., Ingemann, F., Lisker, L., Delattre, P. and Cooper, F.S. (1959)Minimal Rules for Synthesizing Speech, Journal of the Acoustical Society of America,~, 1490 - 1499,

Lisker, L., Cooper, F.S. and Liberman, A.M. (1962) The Use of Experiment in LanguageDescription, Word, ~' 82 - 106.

Mattingly, I.G. (1971) Synthesis By Rule as a Tool for Phonological Research,Language and Speech, ~' 47 - 56.

Slis, I.H. (1971) Rules for the Synthesis of Speech, I.P.O. Annual Progress Report,§.., 28 - 31.

Slis, I.H. and Muller, H.F. (1971) A Computer Programme for Synthesis By Rule,I.P.O. Annual Progress Report, 6. 24 - 28.

3 visual perception

14

RESEARCH ON VISIONH. Bouma

This year's research on vision can be surveyed largely under the same headings aslast year's, although emphasis is shifting towards integrating research on eye sac­cades with research on visual perception. Research on slant perception has decreased.

dynamic properties of human vision

The number of papers on this subject has been extended considerably. A closely inter­

connected series of six articles has now been completed. Three papers have alreadyappeared in Vision Research, whereas the other three are at the press.

Papers I and II (Roufs, 1972 a, b) deal with experimental and theoretical relation­ships between threshold characteristics of flicker and of flashes. Two parameters

characterising "sensitivity" and "inertia" can be deduced from (1) DeLange curvesfor flicker and (2) fIash threshold characteristics. Experiments show the twosensitivity parameters to be simply related, as are the two inertia parameters, over

five decades of background intensity (Figs. 1, 2).

~ Og2 -1 0 1 2 3

log backgr. intensity E

O. subj.I09,oseHJM 0JTHL 0

CII RK 6.E

100 JAJR <>-1.0

..,;

\u0~ ~CIl -1.5 0

.c - .

u e.. e Tc01

.£ _2.0 10

'- 051----+--+--+=,...,..-n---,n+-~. ,./\0~ U-o

~

~1 0 1 2 3 4 I~tdlog backgr. intensity E

Fig. 1. The sensitivity factors S forflicker and F for flashes as afunction of the background inten­sity E, plotted on a double­logarithmic basis (subject H.J.M.).The lower part of the figure showsthe ratio SIF for four subjects.(from Roufs, 1972 a).

Fig. 2. The reciprocal of the cut-offfrequency, llfh for flickerthresholds and the criticalduration T for flash thresholdsas a functIon of the backgroundintensity, plotted on a double­logarithmic scale. The logarithmof the product fbT is also plot­ted. (from Roufs, 1972 a).

Theory based on a few simple deterministic system properties is consistent with theseresults, with the exception of about a factor 2 in the sensitivity ratio. Thisconstant deviation between experiment and theory is attributed to a relatively high

IPO annual progress report 8 /973

15

probability of flicker detection because of the high number of peaks involved, given

the special stochastic properties of the system (paper VI, Roufs, 1974 c).

Integrative properties of the visual system have often been investigated by measuring

intensity thresholds of two identical flashes at variable time intervals (twin

flashes). In paper III (Roufs, 1973) results of this type are related quantitatively

to the DeLange curves and to the threshold characteristics for single flashes.

Curves describing thresholds of twin flashes in relation to flash interval, plotted

on a double-logarithmic base, are found to be isomorphous for different background

levels. From the paper it appears that twin flashes are by no means ideally suited

for measuring integrative properties.

If a combination of a positive and a negative flash (doublet) is used at various time

intervals (paper IV, Roufs, 1974 a), an intriguing difference between experiments and

predictions from DeLange curves is revealed. At levels of 1 and 1200 td, doublet

characteristics are isomorphous, whereas DeLange curves are not -- in fact, the

flicker curves change from a low-pass character at low intensity levels to a band­

pass character at high levels. One way out of the difficulty is to assume that parti­

cularly at low levels, the LF parts of DeLange curves are not representative of the

actual differentiating action. The concepts of considering DeLange curves as composed

of separate LF and HF functions (Fig. 3) is supported by subjective reports at

threshold levels which can be char-

log frequency-Scheme of experimental DeLange curves(solid lines) as a composition of twoseparate curves. The low-frequencybranch would concern brightness varia­tions, experienced as "swell", The high­frequency branch would concern flicker,experienced as "agitation". (from Roufs,1974 a).

>-...:~...'iii swellc:CI)III

Q.ElU

.H'

r

Fig. 3.

acterised as "swell" and ''agitation''

respectively. Once the compound

character of DeLange curves is ac­

cepted, the composing functions can

only be approximated, since these

are not open to direct measurement

as yet. The approximation has been

carried out by computer simulation •

So far, all results concern thres­

hold levels where (quasi-)linear

processes are likely. Ordinary vision

deals with supra-threshold intensi-

ties, however, and the question

arises if the theory can handle

aspects of this as well. The question

is of special relevance to much­

studied problems of visual latency

differences (perception lag) and

of visual reaction times. It appears

that with respect to the general

dependence of these parameters of

intensity differences or to inten­

sity itself, experiment and theory are in reasonable agreement. In particular for

reaction times, however, experimental values at low intensities are substantially

higher than theoretical predictions (Fig. 4). Is perhaps more time required with

weak stimuli for a decision on motor action? (paper V, Roufs, 1974 b).

16

350ms ms

subj.JAJR

300

\100 \

1

log relat ive flash

1501......----L-. L-. l.- _

o

<U~vt-~;:--t--:::J::"""+---+~""'~+----+-----~E

a::....

....c

.Q....uco~

50

F~g. 4. Reaction times for light flashes in relation to flash intensity,taken rel~tive to the threshold of a 400-ms flash. SUbj. JAJRBroken lines have been calculated from DeLange curves forflicker thresholds, measured under the same conditions. Thebroken lines have been shifted together vertically, so as tofit the high intensity reaction times (from Roufs, 1974 b).

visual reading processes

Our research on visual processes involved in ordinary reading has so far concentrated

on two topics:

(a) the nature of eye saccade control and

(b) the recognition of single words in eccentric vision.

As to reading saccades, we have compared ordinary reading with reading in which sub­

sequent retinal images are beyond the control of the reader and within experimental

control (controlled reading). Earlier; we have already reported that both oral and

silent controlled reading are possible(Bouma and de Voogd, 1971), from which we

concluded that in ordinary reading of adults, text recognition and reading saccades

can proceed rather independently of each other. Thus, extent and moment of occurren­

ce of each successive eye saccade can be independent of the ongoing processes of

text recognition, and text recognition can proceed independently of the concurrent

fixation point and of the precise duration of the concurrent eye pause.

Since that report the experiments have been extended (Fig. 5) and a comparison has

been made between reading errors in controlled reading and in ordinary reading to

know if they are of similar type (Bouma and de Voogd, 1974). Classifying reading

errors as omissions, insertions or substitutions, we found in controlled reading a

17

20 pos.o Io£:..._-...,;....:.:.:...l.- ...J..._....J

o

1500 msec.

• •• <) • <) • .~ •••• ••• <J.()()

JA IN

1000 •••• 0 1000 •••• •• <) •• 0

···1 25'5 ••<) • ••

50038·0

500

- Fig. 5. Controlled reading. Externally determined values of shift extent m (letterspaces) and pause duration T (msec) where reading occurred •• Oral reading,~ Oral reading with reading errors above a certain limit, 0 Silent reading.Oblique lines are lines of constant reading speed of the indicated values(letter spaces/sec). Two observers (from Bouma and de Voogd, 1974).

higher fraction of omissions, which turned out to concern mainly short words of many

grammatical classes. In ordinary reading, omissions concerned almost exclusively ar­ticles and conjunctions of 2-3 letters. The difference may point to insufficient vi­

sual perception of short words at the parafoveal limits of the functional visual field

in controlled reading, since in this mode of reading simulated eye saccades were often

larger than saccades in ordinary reading. In other respects we found reading errors

in the two modes of reading of quite similar types: grammaticality is usually left

intact, and in substitutions stimulus and response words are visually similar. This

is taken to show that controlled reading is indeed basically similar to ordinary

reading to the effect that application of conclusions from controlled reading toordinary reading is permissible. The latter conclusion should be demarcated, since

errors of intonation turned out to be somewhat more frequent in controlled reading.

Of course, eye saccades cannot be fully independent of text recognition which they

serve. If the eyes proceeded faster than recognition, portions of the text would of

necessity be missed; if slower, recognition would necessarily retard. For efficient

reading, the point of fixation should proceed on the average just as fast as text

recognition permits, and for bridgeingmomentary differences between the eye and the

brain, a buffer function has to be assumed. For efficient operation, the proceeding

of the eyes over the text should then be under the control of the contents of the

buffer: an empty buffer should lead to an increase in eye speed, whereas a filled

buffer should cause the eyes to slow down. The properties of this buffer or working­

memory need further investigation. As Andriessen and de Voogd (1973) report in the

present issue, successive saccade extents and pause durations are not clearly corre­

lated, which finding is of interest for timing-aspects of the supposed control.

18

Apart from this "global" cognitive control of reading saccades, there should be some

control of individual eye saccades on the guidance of eccentric vision, such as in the

return sweep towards each successive line of print. Finally, reading saccades are sup­

posedly controlled by routine motor programmes.

In parafoveal word recognition l all research problems common to recognition are en­

countered, such as the difficulty of defining the relations between the highly varia­

ble stimuli and the rather conceptualised perception, the influence of redundancies,

several types of bias, and context dependence. In our research we have mainly con­

centrated on a few cues which supposedly serve as intermediates between the stimulus

configurations and the recognised words. Since words are overdetermined, the correct­

ly reported fraction of certain features in response words will usually be higher

than the perceived fraction, owing to other cues aiding the activation of the word

responses. We indicate this effect as "completion" of the cue concerned. For finding

the contribution proper to the word response the measurements should be free of this

completion effect. Therefore, we also used as stimuli unpronounceable letter strings,

from which the observers had to report only the cue under consideration. For initial

and for final letters of words, such an analysis has been carried out (Bouma, 1973,

Fig. 6).

Fig. 6. Word recognition scoresand recognition scoresof initial and finalletters of correspondingletter strings. Allscores averaged overword lengths of 3-6letters. (from Bouma,1973).

-3 -2 -1 0 1 2 3Eccentric stimulus position [degrees]

I 'LG 1-0( r J

u;fl: ' ItJJ~--+i :-1. ~-

I p

~q-2

Dip-0

0-5

At present, word length is also considered in its function as a possible cue to word

recognition. In eccentric vision, the length of letter strings is generally underes­

timated, as reported by Schiepers (1973). It is as yet unclear if the adverseinteractions between adjacent stimuli, which are so dominant in eccentric vision, are

related to this underestimation. Once the separate contributions of a sufficient num­

ber of cues to word recognition have been established, a theoretical frame might be

constructed in which the individual cues can be assigned a proper place.

visual corospicuity

Visual conspicuity is taken here as the stimulus-bound degree of prominence of visual

objects in their background. Thus, we link conspicuity to rather general and constant

analysing properties of the visual system, as opposed to the variable selection pro­

cesses of particular observers, which are partly under their own control. Together,

19

Till abjlel 1: -

Till abjecl 3: U

Till Ibjlel 2: L

Tilt abject 4: 0

F~g. 7. Differences in conspicuity area of four different test objects of the indi­cated configurations (right-hand illustrations) each offered in the samebackground of straight lines (left-hand illustrations). One SUbject. Testobjects in bold lines were reported correctly in a single presentation, thosein thin lines were not. Small central circle: fixation spot (from Engel,1971). these external and internal

factors of attention determine

the actual perceptual selection

out of the mass of visible ob-

in the

foveal vision

jects. The external factor,

conspicuity, was judged the more

fruitful one to investigate. As

a relevant measure of conspicuity,

Engel has proposed the size of

the visual field within which

the object can be noticed in its

background, without pre-knowled­

ge of its position (Fig. 7).

In case the position of the

object is known before the pre­

sentation, a greater area is

generally found, called visibi­

lity area (Fig. 8). Since the

majority of objects, as yet

unnoticed, will be located

seems the proper basis of investigating

Visibility Ani

CaDlP iculty Aru

Attlntian Aru

Pre-knowledge of object position causes theconspicuity area to increase by an adjacentarea in the attended direction (attentionarea). The locus of conspicuity area andattention is the visibility area, where theobject can be seen if its position is knownbeforehand (from Engel, 1971).

eccentrically of fixation, eccentric vision

visual conspicuity.The functions of eccentric vision have been investigated only sparingly

literature. One essential factor in which eccentric vision differs from

Fig. 8.

20

is the large degree of adverse interaction between adjacent stimuli. This is whythe stimulus background is of such importance to conspicuity. We encountered this

type of interference first in our research on reading (Bouma, 1970). Both in the

context of research on slant perception and on conspicuity, these interferences were

found to be specific of certain well-defined stimulus aspects, such as line orienta­

tion (Beerens and Bouma, 1970, Andriessen, Bouma and Beerens, 1971), and luminance

and size (Engel, 1973; this issue, 1974). Consequently, it is at least at these levels

of analysis in the visual system that limits of conspicuity are determined.

If conspicuity is indeed a factor which helps determine visual perception, it is like­

ly to be reflected in eye fixations as well, since selected objects will probably

tend to be fixated. There are a number of recordings in the literature on patterns

of eye saccades in exploring or inspecting pictures, but it is rather unclear how

these eye saccades are guided. We are trying to work our way from visual conspicuity

to eye fixations, and vice versa, in a relatively simple visual search task.

aids for the visually handicapped

In last year's Progress Report, we sketched the general background of our activities

in this field (Bouma, Engel and Melotte, 1972). We argued that in a prosperous

society such as ours, it is necessary that a certain amount of research should be

directed towards bringing useful technological aids under the reach of handicapped

people, and we concentrate on the visually handicapped. The activities should include

construction of potentially useful aids

assessment of their usefulness

further development for meeting the needs of the users as well as of industrial

production

production itself, including its financing, and

distribution.

Existing scattered research tends to concentrate on prototypes and often leaves aside

problems of testing, development, production and distribution. As long as this situ­

ation prevails, we consider it our task to devote the time available for such problems

to bringing indeed a few aids under the reach of the handicapped people involved.

It follows that the progress of 1973 mainly concerns improvements and increased

availability of the two aids of which we constructed prototypes in 1972. These two

were

(a) a closed-circuit TV magnifier and

(b) a embossed-drawing system.(c) as a new project, we have just finished a first laboratory model of a reading

desk, for people with moderate visual acuity to be found mainly in the elderly

category.

(a) The TV magnifier became comm~rcia11y available early in 1973. In cooperation with

the health care group of Philips Nederland BV, we approached at the time ophthal­

mologists as well as organisations and schools of the blind, to inform them on

the apparatus and on the experiences gained in institutes for the blind during

the development phase. This year, we lent our prototype to an ophthalmologist,specialised in Low Vision Aids, and also to an University Eye Clinic. Both users

judged the TV magnifier favourably. A number of enquiries from abroad have been

replied to -- in fact, the TV magnifier can now be ordered in a number of countries.

Production experienced some difficulties, which led to delays, but never~heless

21

some 35 TV magnifiers have been delivered in 1973, several of them financed via

Social Security Laws. We are still working on a few improvements.

55 Children who tried out our prototype in 1972, have been studied more closely in

particular from an ophthalmolugic point of view, in cooperatiDn with the ophthal­

mologic staff of the schools concerned. (Van Veelen, Kinds, Wijnands and Bouma,

1974). Their vision was round about the borderline of visual reading possibilities-­

several of them were in Braille classes. It turned out that 18 of them (33%) were

able to read print of ordinary newspaper size only with the TV magnifier, and not

with their own magnifiers or spectacles. Another 23 (42%) read the print more

conveniently with the TV magnifier than with their own aids. For the remaining

14 (25%), the TV magnifier was of no use, either because they were able to read

rather well without it or because they could not read despite of it. The main

advantage of the TV magnifier over other aids appeared to be not so much a higher

magnification of the retinal image, but rather the increased viewing distance

which the high magnification permits, and, concomitant with it, the naturalviewing situation in which visual contact with the environment remains possible.

(b) In the embossed-drawing system, ordinary writing and drawing on a special sheet

is immediately embossed so that it can be felt with the fingers. The drawing system

met favourable reactions in a number of tests carried out in 1972. In 1973, several

drawing systems were ordered by Dutch schools for the blind and by a few private

blind persons. The production however, encountered some difficulties as to

meeting the required quality. At the time of writing, 75 drawing systems are

almost completed. The drawing system makes use of magnetic clamps for holding the

sheets. Quite a different system of keeping the sheets flat during writing is to

use self-adhesive material under the sheets. However, the regular material is

perhaps slightly toxic, and it is somewhat agres~ive to other materials. Moreover,

we found that it loses its adhesive properties rather quickly. In close conperation

with a chemical laboratory, a composition has been found which we hope is both

non-toxic and stable. This material may prove useful for ordinary embossed writing

and probably also for other applications where objects are to have a fixed position,such as used in certain combination games in arithmetic for blind children. At

the moment the sheets are being tried out. (Fig. 9).

(c) Reading desk. Already for some time we had noticed that reading magnifiers do not

find the enthousiastic reception that their optical quality should permit. We

gOt the impression that a possible reason might be that the available designs are

far from optimal as far as human factors are concerned. In consequence, we have

formulated a number of ergonomic requirements for easy magnifier-reading. Apartfrom safety factors, these include:

large size magnifier of good optical quality. Minimum specular reflection

magnifier always parallel to the text at constant, but adjustable distance

easy shifting of text and/or magnifier relative to each othersufficient illumination

confortable posture when reading

easy handling of the reading desk as a wholeWe have just finished the construction of a suitable prototype (Fig. 10). Our nextstep will be to assess its usefulness in daily practice and to detect functionsto be improved.

22

Fig. 10. Reading Desk ("Lezenaar"). Prototype satisfying a number of ergonomic re­quirements. Construction: Mr. H.E.M. Mel~tte. Lens can be shifted in X-direct­ion, text in Y-direction. Illumination E = 1.500 lux.

Fig. 9. Self-adhesive material which can be used a.o. for keeping flat sheets forembossed drawing.

23

references

Andriessen, J.J., Bouma, H. and Beerens, E.G.J. (1971) Interactions Between Linesin Eccentric Vision: Contrast Threshold and Just Noticeable Difference in Slant,I.P.O. Annual Progress Report, ~, 56 - 60.

Andriessen, J.J. and de Voogd, A.H. (1973) Analysis of Eye Movements in SilentReading, I.P.O. Annual Progress Report, ~, 30 - 35.

Beerens, E.G.J. and Bouma, H. (1970) Orientational Specificity of Line Interactionsin Eccentric Vision, I.P.O. Annual Progress Report, ~, 114 - 119.

Bouma, H. (1970) Interaction Effects in Parafoveal Letter Recognition, Nature, 226,177 - 178.

Bouma, H. (1971) Visual Recognition of Isolated Lower-case Letters, Vision Res.,g, 459 - 474.

Bouma, H. (1973) Visual Interference in the Parafovea1 Recognition of Initial andFinal Letters of Words, Vision Res., 1:2, 757 - 782.

Bouma, H., Engel, F.L. and MEilotte, H.E.M. (1972) Technological Devices for theVisually Handicapped: Gap Between Research Effort and Available Aids, I.P.O.Annual Progress Report, 2, 46 - 54.

Bouma, H. and de Voogd, A.H. (1971) Reading with Fixation Positions of the EyeExperimentally Controlled, I.P.O. Annual Progress Report, ~, 41 - 44.

Bouma, H. and de Voogd, A.H. (1974) On the Control of Eye Saccades in Reading,Vision Res., 1..:1:., in press.

Engel, F.L. (1971) Visual Conspicuity, Directed Attention, and Retinal Locus,Vision Res., g, 563 - 576.

Engel, F.L. and Bos, T.M. (1973) Visual Conspicuity Related to Generalisation ofBackground Regularities, I.P.O. Annual Progress Report, ~, 36 - 42.

Engel, F.L. (1974) Visual Conspicuity and Selective Background Interference inEccentric Vision, Vision Res., in press.

Roufs, J.A.J. (1972a) Dynamic Properties of Vision I. Experimental RelationshipsBetween Flicker and Flash Thresholds, Vision Res., !2, 251 - 278.

Roufs, J.A.J. (1972b) Dynamic Properties of Vision II. Theoretical RelationshipsBetween Flicker and Flash Thresholds, Vision Res., !2, 279 - 292.

Roufs, J.A.J. (1973) Dynamic Properties of Vision III. Twin Flashes, Single Flashesand Flicker Fusion, Vision Res., g, 309 - 323.

Roufs, J.A.J. (1974a) Dynamic Properties of Vision IV. Thresholds of DecrementalFlashes, Incremental Flashes, and Doublets in Relation to Flicker Fusion,Vision Res., in press.

Roufs, J.A.J. (1974b) Dynamic Properties of Vision V. Perception Lag and ReactionTime in Relation to Flicker and Flash Thresholds, Vision Res., in press.

Roufs, J.A.J. (1974c) Dynamic Properties of Vision VI. Stochastic Threshold Fluc­tuations and their Effect on Flash-to-Flicker Sensitivity Ratio, Vision Res.,in press.

Schiepers, C.W.J. (1973) Length Estimation of Letter Strings, I.P.O. Annual ProgressReport, ~, 25 - 29.

van Veelen, A.W.C., Kinds, G.F., Wijnands, H.S. and Bouma, H. (1974) VerkennendOnderzoek naar de TV Loupe als Nieuw Hulpmiddel voor Visueel GehandicapteKinderen bij het Lezen, Ned. Tijdschr. voor Geneeskunde, in press.

24

LENGTH ESTIMATION OF LETTER STRINGSC.W.J. Schiepers

introduction

~eneraZ

In normal reading, information from a text is picked up by the eyes, processed andpassed on to the brain. The eyes move with quick jumps along a line. Only during the

eye-pauses (200 - 500 ms) is there an intake of information consisting of letters and

words that are combined into sentences. We are interested in the recognition proces­ses and will explore what stimulus properties are relevant to these processes.

Bouma (1970, 1973) has studied some details of the functional visual field, i.e. thearea round the point of fixation within which a certain stimulus property can be

seen during a single eye- pause. This area is larger than the fovea because para­foveal stimuli can also be recognised.In reading relevant stimulus material ispresent in the parafoveal region.

Words

We restrict ourselves to single words and try to discover what kind of global attri­

butes, i.e. more word-bounded rather than letter-bounded attributes, are important in

the word recognition process. Of these the length, i.e. the number of letters, andthe shape, i.e. the total picture of the word constructed by the letters, are twopossible candidates.

Erdmann and Dodge (1898) argue strongly in favour of a general word shape, by which

we recognise the word, and not a stringing together of its letters.

Length

In the present experiments we examine the part word length plays. In parafoveal wordrecognition experiments Bouma, (1973) reports that scores of correctly recognised

words decrease with both an increase in stimulus length and an increase in eccentric­ity of presentation. In addition, there is a left-right (visual field) difference,viz. recognition scores are somewhat higher in the right-hand field, which increaseswith increasing stimulus length.

In exploratory word recognition experiments, Bouma and Van Rens (1970) found that thelength distributions of the confusions, i.e. not correctly recognised words, werenot dependent on eccentricity of presentation. In normal words the influence oflength alone cannot be traced. Words have many more properties by which they can berecognised; evidently, their reported length is then correct.Therefore, we have chosen the way from rather simple stimuli (strings of identicalletters) via nonsense letter combinations to words. We assume that an estimation of

the length of a letter string with the same contour as some particular word, is ameasure of the attribute 'length' in recognising that word.Here we restrict ourselves to:

a. strings of identical letters

b. strings of arbitrarily chosen letters.

method

Stimu Zi

Using an IBM-72 typewriter, the letter strings, i.e. a number of letters Ixl without

IPO annual progress report 8 /973

25

spaces, were typed on white paper. The lettertype was Courier-la, of which the

height of ascenders and descenders is 2.70 mm and that of short letters 1.95 mm. The

letterwidth varied from 1.4 mm (j) to 2.5 mm (m, w). All letterspacings measured

between centres were 2.55 mm.

We define as nominal eccentricity of a stimulus the distance from the fixation point

to the middle of the nearest letter, in degrees of visual angle. There were also

stimuli centred round the fixation point, which we shall call 'foveal stimuli'.

Presentation

The stimuli were presented by a two-channel tachistoscope. In the rest field of about

30 x 30 em, illuminated by white light, luminance 150 cd/m 2, a fixation point was

placed. This field was replaced during 100 ms by a similar one in which a stimulus

was present. The stimuli were randomly presented either left or right of the fixa­

tion point. The subject depressed a microswitch to initiate a presentation. Viewing

distance was 57 cm, at which 1 cm (4 letters) corresponded to an angle of 1°.Vision was binocular.

Subjects

Seven students of the Eindhoven University of Technology acted as subjects. Their

ages were 19 - 25, they all had adequate vision (foveal acuity 1.25 - 2) and were

all right-handed. They were naive as to the aim of the experiments.

Instruction

Subjects were asked to report the number of letters they had seen. In doubtful cases,

two responses were allowed, e.g. 4 or 5. In those cases each response category got

a score of ~.

results

L= 4

l= 5

l= 3

5

L= 2

•

4

•

•

3

•

•

2

p •

Lee (c'egrees)

.<;<--0_-0"="--0-----0-0-- l~ 1

-0 0-1

•-2

•

-3

•

-4

•

·9

.8

·7

·6

·5

-4

.3

·2

.1

0-5

Averaged scores of correctly reported length in relation to eccentricity, as attained

by the seven subjects, are given in Fig. 1. From the fixation point outwards the

scores decrease and become roughly const~nt for large eccentricities. This was also

the case for the other stimulus lengths.1.()

Fig. 1. Averaged scores p of correctly reported lengths. The points between _0° and0° are the scores of the foveal stimuli. Number of stimuli per point:n = 35 (1 = 1, 2), n = 105 (1 = 3, 4, 5), at ¢ = + 1. 75° three times asmuch. The lines have been drawn by hand and try to indicate the trend of thes core c'

.751----~--- --~----

26

It is somewhat surprising that the scores do not continually decrease with eccentric­

ity, which is often the case in visual recognition tasks (acuity, contrast, word

perception). In Fig. 2 histograms are given of the length distributions of the re­

sponses; left and right scores are averaged, there being no significant differences.

There is a general tendency to underestimate the length and the more so as the stim­

ulus length increases ( Stimulus lengths 1 and 2 form a separate group, which we do

not consider).

l.O°r---------------------------------:l-e.,-tt,.-e-r-s-/.,-x"""/-.,= ~ 1,75°

.50

.25

OL-----I.-'-:::-"'-_~I.....L._:_'_-....l:::1....l~-.....J.......L...l...."=--...c:L....L....LJ=--ClLl.....L...J..........L...1...J.....L.....I::::L-.......Ll...LI....J.-.J4 5 6 7 8 5 6 7 8 9 6 7 8 9 10

rig. 2. Length distributions of the responses. Left-right scores are averaged. Largesymbols: stimulus lengths. Number of stimuli: n = 630 (1 = 3, 4, 5, 6, 7),n = 210 (1 = 8, 9, 10)

To describe this underestimation in a simple way, we introduce a new parameter:

relative

mI

perceived length A

mean of reported lengthsstimulus length

Fig. 3 shows this relative perceived length A in relation to eccentricity for the

various stimulus lengths. If, for a while, we exclude 1 1, 2 and the central area

( I¢I > 2°), A is roughly constant. For 1 > 5,A and the variance of A are not

significantly different, i.e. the length distributions are equal. We can conclude

that the perceived length of letter strings can adequately be described by A, which

is - not dependent on stimulus length (1 ~ 3)

- not dependent on eccentricity of presentation I ¢ I > 1 0) •

In the foveal area A is somewhat higher for small stimulus lengths (1 < 5). Just

near the fovea most stimulus lengths have a lower A.We found for the letters Ix/:

A = 0.85 + 0.01 (1 ~ 3,1¢1 > 1°, 95% can£.)

We also used strings of letters Idl and lei at ¢ = + 1. 75°. These letters show

comparable pictures; their lengths are systematically underestimated. For both

letters at ¢ = + 1. 75°:

0.78 + 0.01 (1 ~ 3, 95% con£.)

Fig. 4 represents histograms of the length distributions of the responses of strings

of arbitrarily chosen letters, averaged over seven subjects. The distributions are

comparable with those of the letters Ix/. We can again adequately describe the

estimated lengths by a A, which is not dependent on the stimulus lengths (1 > 3) and

eccentricity. It would, therefore, seem that the extensions of the letters have no

influence.

After each session the subject was asked to describe what he or she had done. They

all reported that for larger stimuli most letters were indistinguishable from each

other--they saw a 'gray stripe' of which they estimated the length.

27

1.1 0r-........--....----r-----,.----.----r--,---.......--........--....---.....,,.......-.......--,

1.0

.95

.90

21

2

21 21 21 12

2

2

.75

.70

-5 -4 -3 -2 -1 -0 0 2 3 5

F~g. 3. Relative perceived length A. Symbols indicate the stimulus length, 0 for1 = 10. Symbols correspond to the middle of the stimulus (for surveyability).

let tcrcombinilbi OilS

cp == .:. 1 ,7S

r-

'-------

'Il h Jl.--., nm-I I I" 1 I I JII I rl "I

lcttcrcomlJi nol ionscp == t 2, 7So

->---

K h --, I--, nrllrLL r L I" L J L J rh

1.00

.75

.50

.25

o

1.00

.75

.50

.25

o3~567 ~5678 45678956789 67891011

7 8 910

7 8 91011 12

~ig. 4. Length distributions of the responses.Stimuli: Strings of arbitrarily chosen letters. Lar3e symbols: stimuluslengths. Number of stimuli: n = 105 (1 = 3), n = 140 (1 = 4, 5), n = 175(1 = 6, 7), n = 35 (1 = 8, 9, 10).

1-: estimated lengthL: stimulus length

k, n: constants

28

discussion

Estimation of the length of lines can adequately be described by the psychophysical

law of Stevens:

1- = k Ln

with constants k and n of about 1 (Stevens and Galanter 1957, Baird 1970). Our re­

sults, too, fit quite well into the equation with an exponent n = 1 and a constant

k of about 0.8 in the parafoveal area. In the central visual field, however, the

situation cannot be described so easily. An increase in A in foveal vision seems

reasonable, for there are more details visible than in parafoveal vision. We do not

have a suitable explanation for the dip just next the fovea.

In word recognition experiments, recognition scores show a left-right difference

(Mishkin and Forgays 1952, Bouma 1973). This difference increases with increasing

stimulus length (Bouma 1973). Our results show no left-right field difference in the

parafovea. The subjects said that the letters were indistinguishable, thus parafoveal

interference as reported by Bouma (1970) plays an important part. Bouma (1973) pro­

poses that the spatial extent of this interference (in words) is larger in the left

visual field than in the right. Therefore, if we take interference as an explanation

of the underestimation, it cannot be the same type of masking as the interference inreal words.

Words have many more attributes that may be important in the recognition process. The

aim here was limited to extracting the influence of word length. This cannot be a

very strong one, for recognition scores of words do depend on stimulus length and

eccentricity. Letter strings do not show such dependence; therefore, other word

attributes must be responsible for the decay in the recognition scores.

summary

Length estimation of letter strings has been investigated in relation to stimulus

string length and eccentricity of presentation. Surprisingly, correct scores are not

dependent on eccentricity for angles I~l > 2°.Length is sytematically underestimated, average of judged length is approx. 85% of

stimulus. length.

Results are considered to reflect the role of string length in word perception.

references

Baird, J.C. (1970) The Psychophysical Analysis of Visual Space, London, Pergamon Press.

Bouma, H. (1970) Interaction Effects in Parafoveal Letter Recognition, Nature, 226,177 - 178.

Bouma, H. and Van Rens, A. (1970) Reading Processes: on the Recognition of SingleWords in Eccentric Vision, I.P.O. Annual Progress Report, ~, 99 - 106.

Bouma, H. (1973) Visual Interference in the Parafoveal Recognition of Initial andFinal Letters of Words, Vision Research, l1, 767 - 782.

Erdmann, B. und Dodge, R. (1898) Psychologische Untersuchungen liber das Lesen aufExperimenteller Grundlage, Niemeyer, Halle a.S.

Mishkin, M. and Forgays, D.G. (1952) Word Recognition as a Function of Retinal Locus,J. Exp. Psychol., ~, 43 - 48.

Stevens, 5.5. and Galanter, E.H. (1957) Ratio Scales and Category Scales for aDozen Perceptual Continua, J. Exp. Psychol., ~, 377 - 411.

29

ANALYSIS OF EYE MOVEMENT PATTERNS IN SILENT READINGJ.J. Andriessen and A.H. de Voogd

introduction

As is well-known in reading, the eye movement pattern mainly consists of quick hor­

izontal saccades along the lines of print, separated by longer eye-pauses in which

the reading proper takes place (Woodworth and Schlosberg, 1954).

In the present contribution we discuss the analysis of such eye movement patterns

during ordinary silent reading, with two aims:

1. to characterise the statistics of fixation durations and extents of eye saccades

by means of histograms;

2. to find properties of the motor control of the eyes from possible correlations

between successive fixation durations and successive saccade extents.

As to the latter, it might be assumed that each successive fixation is programmed in

connection with the recognition of text during the preceding fixational pause. The

feed-back system needed for such control would require small time constants and is,

in fact, unlikely (Bouma and De Voogd, 1974). Therefore, any significant correlation

would rather seem to reveal properties of motor programmes operating on a routine

basis, independent of text recognition. As an example, a small saccade might general­

ly be fOllowed by a large one, in which case the average of two successive values

would result in a rather constant value. The same might hold for fixational pauses,

a long one being followed by a short one, and vice versa. We have so analysed the

recorded eye movements that the degree of correlation can be shown graphically.

method

The text to be read was placed vertically at 57 em distance. The field in which the

text was presented subtended 20 0 visual angle horizontally by 150 vertically.

Luminance was about 300 cd/m 2• The text was black typescript on white paper. Type

face was "Courier" (IBM 72, 10 pitch). The average letter width was about 0.20 0

visual angle. The typewriter space was about 0.25 0, which is henceforth defined.as

"one-letter position" or "position". The text consisted of 9 lines of about 65 letter

positions each. The lines of print were 1.5 0 apart. Six texts of equal length were

used in the experiments. They were all simple texts taken from a book in the Dutch

language. The instruction given to each observer was to read each text silently at

his own speed. There was a break between each two trials.The eye movements were measured by means of the cornea reflection technique

(Television Eye Marker, Mackworth, 1958; Schumm 1967). Briefly, this technique sup­

plies an electronically mixed signal of two TV cameras, one for the corneal reflect­

ion and one focussed on the text to be read. It gives TV pictures which are only

visible to the experimenter and in which the point the observer fixates is represen­

ted by a bright spot superimposed on the image of the text. The TV images are

recorded on video-tape.

From the TV images the successive fixation points along the lines of print were

determined, as well as the extent of each saccade and the duration of each fixation

pause. The latter was determined by counting the number of TV frames involved in a

pause (1 TV frame ~ 20 ms). The accuracy of the cornea reflection method can be

strongly affected by head movements. Therefore, head movements were kept to a minimum

by using a bite-board. But, head movements cannot be avoided totally, the result

IPO annual progress report 8 1973

30

being that recorded eye movement patterns may be shifted with respect to the text.

So, the absolute position of the fixation point in the text may be less accurate

than its relative position. The relative accuracy of the position is roughly ~ 1

letter position per ZO saccade extent ( 8 letter positions). The fixation duration

has an experimental error of ~ ZO ms (~ 1 TV frame).

The experiments were carried out by 7 observers with adequate vision. They used

binocular vision.

results

The results of the analysis have been pooled for the 7 observers and the 6 texts.

They are summarised in Figs. 1-6 as frequency distributions. The vertical axes give

the relative frequencies of occurrence in per cent.

Fig. 1 shows the histogram of the fixation durations. The fixation durations have a

slightly skew distribution with an average of about ZOO ms and a standard deviation

of about 60 ms. Fig. Z represents the histogram of the saccade extents. There

appear to be three separate distributions. The extents of the forward saccades have

also a skew distribution, the average being about 8 letter positions (2° visual

angle) with a standard deviation of about 3 positions. There are two smaller fre­

quency distributions, one of the return sweeps (-65 to -46 positions) and one of

the regressions (-10 to 0 positions).

In Fig. 3 the relation between successive saccade extents is shown graphically. The

preceding saccades were classified into 4 categories: small (1 to 5 positions),

medium (6 to 10 positions), large (16 to Z5 positions), and the return sweeps

(-65 to -46 positions). Each curve in Fig. 3 depicts the conditional distribution

of saccade extents after the saccade with the specified extent (Saccades with an

extent larger than -10 positions have been omitted).

If conditional distributions coincide, this means that, irrespective of the pre­

ceding saccade, the same distribution of successive saccade extents is found. The

correlation between successive saccade extents is then low or zero. As will be

seen from Fig. 3, there is hardly any correlation between successive saccade extents.

Only after return sweeps does the relative frequency of regressions (-10 to 0

positions) increase. It appears, in fact, that the return sweep is often too small

(Fig. Z, left part) as a result of which it is rather freque71tly followed by a

regression. There is a tendency that after a large forward saccade the relative

frequency of regressions increases as well. In Fig. 4, the relation between

successive fixation durations is shown in a corresponding way as in Fig. 3. Here,

the preceding fixation durations were classified into 3 categories: "short" (20 ­100 ms), "medium" (160 - 240 ms) and "long" (3Z0 - 500 ms). The curves appear to

coincide fairly well, which indicates that between successive fixation durations no

correlation occurs.

Fig. 5 illustrates the relationship between fixation duration and the extent of the

saccade following. The durations of the preceding fixation pause were classified into

2 categories : "medium" (160 - Z40 ms) and "long" (320 - 500 ms). For the category

termed "short" (20 - 100 ms) the number of data was too small to get a reliable

histogram. It turned out that the saccade extent is not influenced by the preceding

fixation duration.From Fig. 6 may be seen the relation between saccade extent and the duration of the

fixation pause following. The preceding saccades were classified into the same 4

31

16 .,.

12

,2

°0~--::10~0--:2~0::0--:3~0':'"0--'~00"":;=-C5:1:100--600 ms

fixation duration

Fig. 1. Fixation durations in silent reading (N = 3499).

10 .,.N=3457

8>-u&i 6:l0-

£ 4

Q/...2

0·70 -'0 -30 -20 -10 30

saccade extent (pas.)

Fig. 2. Saccade extents in silent reading (N 3457).

32

0--0 20 -100 mi. n .155

A- -A. 160·240 mi. ".2107

A--------6. 320-500 mi. n. 264

600ms200 300 400 500fixation duration

20 % 20 %

0--0 1-5 pos.na48.4

18 1><-1>. 6-10 pos. n.1227 186----1'. 16-25 pos. n. 116

16+.,.+ -65-46 pOS. n. 323

16

14 + 14

12 12>.

>. u

glO c+ OJ 10

Q/ ::l::l

~0-

0- OJOJ 8 L- 8~ -Q)

....;OJ

L- 6 I- 6I I

4 ~", 4

2 +1- 12A 1+

0 0-10 -5 0 5 10 15 20 25 30 0

saccade extent (pos.>

Fig. 3. Conditional frequency distri­butions of saccade extents. Thepreceding saccades were classi­fied into 4 categories as indi­cated in the diagram.

Fig. 4. Conditional frequency distributionsof fixation durations. The prece­ding fixation durations were clas­sified into 3 categories, as indi­cated in the diagram.

0---0 1 - 5 pos. ",,624

A--;A, 6 -tOpos. n.1384

~ 16 -2Spos. n.125

+ + -65 - 46pos. n.324

20%A--A 160-240 ms. n.1943

18 t>---{', 320-500 mi. n_U6

16

14

12

>.u 10cOJ::l0- 8OJ~

Q) 6L-

4

2

0~ ~ 0 5 ro ~ 20 25 ~

saccade extent (pos.>

20 %

18

16

14

12

>.uCOJ::l0-Q/

~....;OJI-

++

100 200 300 400 500fixation duration

600ms

Fig. 5. Conditional frequency distri­butions of saccade extents. Thepreceding fixation durationswere classified into 2 catego­ries, as indicated in the dia­gram.

Fig. 6. Conditional frequency distributionsof fixation durations. The prece­ding saccades were ~lassified into4 categories, as indicated in thediagram.

33

categories as for Fig. 3.If the preceding saccade is small (1 to 5 positions), or in the case of a returnsweep, the fixation pause following has more frequently a short duration than in the

case of a medium forward saccade (6 to 10 positions), but the effect is slight.

discussion

From the above results it appears that, at least during silent reading, the eye

movement control mechanism follows a programme of more or less random saccade extentsand fixation durations with a mean of about 8 letter positions and 200 ms respec­

tively. Therefore, we find no evidence of a precise programming of every single

saccade extent or fixation duration. A small saccade is not necessarily followed by

a large one, nor is a short fixation pause necessarily followed by a long one. Onlya few visual cues from the text seem to exert influence on the control, as forinstance, the end of line for carrying out the return sweep. We have found that

small forward saccades and return sweeps are fairly frequently followed by relative­ly short fixation pauses, which phenomenon might be caused by what may be called a

"restarting procedure". In an attempt to explain this the case of a return sweep is

taken in which the beginning of the line is rarely reached at once. It is nowassumed that the fixation pause is then shortened in order to find the intended fixa­tion point by an additional saccade. Perhaps parafoveal vision plays a part in the

process. The numerical values of saccade extent (in letter positions) and fixationduration correspond fairly well with data of Cunitz and Steinman (1969) and witholder data, for instance those reported by Huey(1908) and Woodworth and Schlosberg

(1954). It appears from this literature that it is preferable to express saccadeextent in letter positions rather than in visual angle, as may follow from Huey(p. 29):

" Doubling the reading distance did not appreciably lessen the number of pauses

per line. Of course this means that the angle of each eye-movement grows smalleras the book recedes ".

The nature of saccade control in reading is discussed in more detail in Bouma and DeVoogd (1974).

summary

The cornea reflection technique has been used to record the eye movement patterns

during silent reading. The data comprised the sequence of fixation pauses along thelines of print, the extents of the saccades (normally forward, regressions andreturn sweeps) and the durations of the fixation pauses. The correlation between thedurations of successive fixations and between successive saccade extents has been

studied; likewise the correlation between fixation duration and saccade extent, andvice versa.

It follows from the results that only in a few cases might some correlation exist.

Generally, it turns out that the proceeding of the eyes across the text follows aprogramme of more or less random saccade extents and fixation durations, with a

mean of 8 letter positions and 200 ms respectively. Only a few visual cues seem to

influence the programme, for instance, the end of line for carrying out the returnsweep.

34

referencesBouma, H. and De Voogd, A.H. (1974) On the Control of Eye Saccades in Reading,

Vision Research, in the Press.

Cunitz, R.J. and Steinman, R.M. (1969) Comparison of Saccadic Eye Movements duringFixation and Reading, Vis. Res.,,2., 683 - 693.

Huey, B.B. (1908) The Psychology and Pedagogy of Reading, MacMillan Cy., M.I.T.Press, 1968.

Mackworth,J.F. and Mackworth, N.H. (1958) Eye Fixations Recorded on Changing VisualScenes by the Television Eye Marker, J. Opt. Soc. Amer., ~, 439 - 445.

Schumm, C. (1967) Measuring Eye Movements by Means of Corneal Reflection, I.P.O.Annual Progress Report, !' 123 - 129.

Woodworth, R.S. and Schlosberg, H. (1954) Experimental Psychology, 504 - 510,Henry Holt, New York.

35

VISUAL CONSPICUITY RELATED TO GENERALISATION OF BACKGROUNDREGULARITIESF.L. Engel and T.M. Bos

introduction

As a measure of visual conspicuity the size of the so-called conspicuity area has

been proposed (Engel, 1969). This area corresponds to the retinal field in which one

is able to discover the object involved from its background in a brief exposure,

allowing of a single eye fixation only. The conspicuity area has to be distinguished

from the retinal field in which one can perceive the relevant object when its eccen­

tric location is indicated in advance. For complex backgrounds these fields are

generally larger than the corresponding conspicuity areas, which demonstrates the

influence of directed attention (Engel, 1970).

This paper describes some experiments concerning the perceptibility in eccentric

vision of a test disc in a background of randomly located luminous discs, with

foreknowledge of test disc location. The experiments link up with earlier conspicuity

area determinations, performed with a similar test object-background combination, in

relation to test disc diameter and luminance (Engel and Bos, 1971) and in dependen-

ce On test disc colour (Engel, Ligtenbarg and Bos, 1972). Since perceptibility is a

requirement for conspicuitY,the test disc perceptibility data may supply more insight

into the subsequent selective reduction processes in the visual system. :)ut of the

available visual informatio~ for instance, a first selection is made through eye­

movements in association with a limited visual field. Attention, too, may be consider­

ed to perform a certain selective action; actually, we assumed visual conspicuity

to be an external determiner of selective attention (Engel, 1969).

As already observable in Fig. 1, the background discs impede the perceptibility of

the test disc in eccentric vision. As will follow from the experiments to be des­

cribed, even the detectability of the presence of the test disc at a location in­

dicated beforehand appears to be affected by these background discs. Since this

interference is found to be largest for a test disc similar to the background discs,

it points to some generalisation process, by which similarities in the stimulus

pattern, such as, in our case, similar objects, are segregated as a group, so that

dissimilarities are relatively emphasised. The experiments to be described are part

of a more extensive publication in print (Engel, 1973).

experimental

If information on test object location is supplied in advance, the question of per­

ceptibility of the object divides into a question on the possibility of detecting

the object and a question on the possibility of discriminating the test object from

the background objects. Actually the first point constitutes a requisite for the

second, just as the second point must be satisfied for the test object to attract

attention by conspicuity. In the discussion we return to an exception to the latter,

namely to conspicuity through relative isolation of the test object from the back­ground objects.

In accordance with the points mentioned above let us distinguish the following visualfields:

1. The visibility area, in which the presence of the test object can be perceived

at the eccentric position indicated beforehand, during a brief presentation

/PO annual progress report 8 /973

36

(SO msec) of the object­background combination.

2. The discrimination area.

in which - with fore­

knowledge of the posi­tion - moreover a dif­

ference relative to thebackground objects can

be discerned.3. The conspicuity area.

in which - without fore­knowledge of the position ­the observer discoversthe test object in the

background during the

brief exposure.

method

Complex background

• ••• •••••• •••• •••••• •• • ••••• •• •• •• •• • ••• ••• • •• • •••• •• • •• •• • •.... . '... . .• ••••• ••• •• •• • •• ••• ••• • •• • •• •• •• ••• • • •••• ••• • • • •• •·.. : : : .• ••• •• •• •• ••••••••• •••• • • •••• •• • •••••

• • •• •••• •• • •••••••• ••••• ••••••••• •• ••• • •• •• •• • ••.. :. . .• •• • •• ••• •••• • ••... . : ..

• •••• •• • • •• • ••• •••••• ••• •• •• •••• ••• ••••• • •••••••• ••••••• • • ••••••• • ••• • • • • •• •• •• • •••

The test disc in a plain (above) and a structured(below) background respectively. As can be judgedby eccentric fixation, the pe~ceptibility of thetest disc is decreased through the addition ofbackground diacs.

Fig. ;l..

We used slides with thedisc pattern of Fig. 1.These slides differed onlyin the diameter or bright­ness of the test disc,

whose position in thebackground pattern was the

same for each slide. Theobject-background combina­tion was projected on atranslucent screen (100 x80 cm) by means of a two­channel tachistoscope. Itwas possible to position

the test disc on the screenby shifting the slide. Since

the slide was only partlyprojected, the backgroundalways filled the screencompletely, each time dis­playing some 130 discs.

The diameter Do of the background discs corresponded to 3.5 0 visual angle at 57 cmviewing distance, while their luminance Lo equalled 33 cd/m 2

• The luminance Lb ofthe dark part of the background was 0.65 cd/m 2 • The luminance of the rest field pre­sented between the 80-msec stimulus exposures via the second channel of the tachis­toscope, equalled the average luminance (4.5 cd/m 2 ) of the stimulus field. A smallfixation dot was continuously supplied in the centre of the screen. The 57-cmviewing distance was maintained with the help of a head-rest, ensuring the right eyeof th~ observer to be in front of the fixation spot. The left eye of the observer

37

F.E.

0 ........._ .................._ .........oo 2 4 6i -0

DO

0 ........._ .................._ ........

o 2 4 60

1- 0Do

2

F.E.

o0.65

1Lb

10

2

O _ ~

Q65

iLb

10

Fig. 2. The size of the visibility area (R ), the discrimination area(R

d) and the conspicuity area (R )v in relation to test disc

diameter (D) and luminance (L), for the observers TB and FE.

38

was shielded with an eye-cap.

Plain background

A large translucent semi-sphere (90 ern in dia.) was available for measuring the

visual fields of the test disc without the background discs. The luminance of the

semi-sphere (0.65 cd/m 2) was equal to that of the dark part of the complex back­

ground. The test disc could be displayed from outside the semi-sphere at any desired

location, also for 80-ms periods. The semi-sphere was also here provided with a fixa­tion dot and a head rest.

Area determination

The borders of the three areas were determined relatively to the central fixation

spot in 8 directions (viz. horizontal, vertical and diagonal).

The visibility area was determined by the method of limits. For that purpose, before

each exposure, the stimulus pattern was shifted outwards and inwards in small steps

along the meridians successively. The observer reported if he perceived the presenceof the test disc.

The discrimination area was determined in the same way; here, however, the observer

reported whether he saw the difference between the test disc and the background

discs.

The conspicuity area was determined by positioning the test disc on one of the meri­

dians at random, while after each exposure the observer reported if he had discovered

the test disc by indicating the supposed location on the screen by means of an electric

torch. The averaged border eccentricity ~ was called the "size" of the visual field

concerned.

results

Fig. 2 gives the data referring to the experiments with the test disc in the complex

background. The data are related to test disc luminance and diameter. Two experienced

observers participated. Their results agree reasonably well with each other. All

field sizes are of course nil for a test disc diameter D=o and for a test disc lumi­nance L = Lb , the luminance of the dark part of the complex background. Moreover,

the size of the discrimination area and, consequently, of the conspicuity area,becomes zero for a test disc identical with the background discs (L = Lo and D = Do)'Note that the size of the visibility area (R ) is non-zero there, since it is still

vpossible to detect the presence of such a test disc with foreknowledge of location.

However, the size of the visibility area shows a decrease for D and L approaching

Do and Lo respectively, which is not self-evident. The existence of this effect hasbeen corroborated by other threshold measurements (Engel, 1973).

The R -data for the plain background are depicted in Fig. 3, together with thev

corresponding complex background results from Fig. 2. The interfering influence of

the background discs becomes evident in Fig. 3.

discussion

The extent of the monocular visual field has already been investigated thoroughly

by means of perimetry. Our plain background results presented in Fig. 3 agree with

the perimetric data supplied in the literature (e.g. Ferree and Rand, 1931).

The difference between the complex and plain background data in Fig. 3 shows the

interfering influence of the background discs. These interferences cannot be comple-

39

52

•.--.-.---.' -+-+-+.+,

.=l.B

/I += F.E

'/-..~~~

20 ~,I,, °=°00.'...._ ................_ ..........

Q65

flb

40

60

Rv

f

4

r -0DO

2

A= 1.B

~---y-

0 ............._ ......_ ..........o

60

E'ig. 3. The size R of the visibility area in relation to test disc diameter D andluminance L. The lower functions are obtained with the complex background,the upper ones with the plain background. The relative minima at D = D andL = Lo ' demonstrate the specificity of the background interferences. 0

tely understood on the basis of visual acuity data, as they occur over too large

retinal distances (the discs are separated by more than 8° visual angle) and, moreover,

apparently include a diameter selective component and a luminance selective com­

ponent. Actually, we assume that R increases again for L > L just as it does forv 0

D > D . In addition to a selective decrease in R for the complex background relativeo vto the plain background, a more general decrease of R can be observed in Fig. 3. At

vleast part of this general decrease can be explained by realising that the luminance

of the test disc equals that of the background discs for Rv in relation to test disc

diameter D, so that in that case the luminance selective interference component is

maximum and constant. This argumentation also applies to Rv in relation to test

disc luminance L, where the diameter selective interference component is maximum

and constant, since the diameter of the test disc is here equal to that of the back­

ground discs.

It may be of interest to note that orientation selective interferences in eccentric

vision have also been established (Andriessen, Bouma and Beerens, 1971). If we

assume that the interferences reflect selectively of interacting feature units, our

interferences constitute indications of the existence of size and luminance specific

units.

In the literature,an increasing amount of evidence on the existence of size and

40

orientation selective processes has already been supplied. However, to our knowledge

no luminance selective effect has been reported as yet. But the interferences mayoccur at a level where luminance and size specificity are combined. This possibility

approaches closely the "Gnostic units" proposed by Konorsky (1967).

An attractive explanation of the origin of these selective interferences is that

they result from lateral inhibition effects between similarly tuned feature units.

This, indeed, leads to generalisation of similar objects in the stimulus pattern and

to relative distinction of different ones. Since lateral inhibition tends to decrease

with lateral distance, relative isolation of an object will also increase relatively

the excitation of the corresponding units. This influence on the size of the conspi­

cuity area has, indeed, been reported in an unpublished work of Beekmans and

Van Der Laarschot (1971). Disc density also influenced the size of the conspicuity

area (unpublished work of De Jong and Duijnhouwer, 1972). Our results may have some

relation to perceptual grouping, or so-called "Gestalt effects" (Wertheimer, 1923),

occurring in dependence on spatial proximity and object similarity (Beck, 1972).

The size and luminance selective interferences imply emphasis of figure-ground

relationships already at the pre-attentive visibility stage, in line with ideas

proposed by Neisser (1967). This also fits in with the tentative flow-diagram pro­

posed earlier (Engel 1969, 1971), see Fig. 4.

AttentionStimuli

Memory,

Consciousness,

etc.

Expectation

Response

Fig. 4. The tentative flow-diagram of visual information processing, proposed earlier(Engel, 1969). A certain selection already occurs at the visibility stagethrough eye movements and selective interference. A further reduction isaccomplished by selective attention.

Conspicuity as an external determiner of selective attention (Engel, 1971) may thenbe conceived as a relatively increased activity in certain feature units, whose

dominance may consequently trigger selective attention.

summary

Experiments are described concerning the perceptibility and conspicuity of a test

disc in eccentric vision, in relation to its diameter and luminance, when embedded

in a background of equally sized, randomly located, luminous discs. These discs

appear to exercise selective interference on the perceptibility of the test disc,

as appeared from the size of the visual fields in which the test disc was perceptible

with foreknowledge of location. Since these interferences are largest for a test

disc similar to the background discs, they point to some generalisation process by

which irregularities in the stimulus pattern, such as difference in diameter or

luminance, are relatively emphasised with respect to similarities. An indication is

given of the relation between this pre-attentive selection and conspicuity. The

investigations described are part of a more extensive publication which is in print.

41

references

Andriessen, J.J., Bouma, H. and Beerens, E.G.J. (1971)Interactions between Lines inEccentric Vision: Contrast Threshold and Just Noticeable Difference in Slant,I.P.O. Annual Progress Report, ~' 56 - 60.

Beck, J. (1972) Similarity Grouping and Peripheral Discriminability under Uncer­tainty, American Journal of Psychology, ~' 1 - 19.

Beekmans, A. and Van Der Laarschot, F. (1971) Visuele Opvallendheid als Funktie vande Achtergrond, I.P.O. -R. 225.

Engel, F.L. (1969) In Search of Conspicuity; Attempts to Measure and ComprehendVisual Conspicuity, I.P.O. Annual Progress Report, ~' 89 - 95.

Engel, F.L. (1970) Visual Conspicuity and Directed Attention, I.P.O. Annual ProgressReport, ~' 123 - 128.

Engel, F.L. (1971) Visual Conspicuity, Directed Attention and Retinal Locus,Vision Research, !l, 563 - 576.

Engel, F.L. and Bos, T.M. (1971) Size and Luminance Contrast in Visual Conspicuity,I.P.O. Annual Progress Report, ~' 45 - 48.

Engel, r.L., Ligtenbarg, H.G. and Bos, T.M. (1972) Conspicuity and Colour,LP.O.Annual Progress Report, 2, 55 - 62.

Engel, F.L. (1973) Visual Conspicuity and Selective Background Interference inEccentric Vision, I.P.O.-MS 239 II, accepted for publication in Vision Research.

Ferree, C.E. and Rand, G. (1931) The Effect of Relation to Background on the Sizeand Shape of the Form Field for Stimuli of Different Sizes, Am. J. Ophthalmol.,~' 1018 - 1029.

De Jong, J. and Duijnhouwer, W. (1972) Opvallendheidsgebied en Achtergrondsdichtheid;Oriinterende Experimenten, I.P.O. - R 247.

Konorsky, J. (1967) Integrative.Activity of the Brain; an Interdisciplinary Approach,The University of Chicago Press, Chicago, London, 73 - 110.

Neisser, U. (1967) Cognitive Psychology, Appleton Century Crofts, New York, 86.

Wertheimer, M. (1923) Untersuchungen zur Lehre von der Gestalt II, Psychol. Forsch.,~' 301 - 350.

4 perceptual and motor skills

42

PERCEPTUAL AND MOTOR SKILLS

W.G. Koster

introduction

Rearrangement of the group and the prospect that three members will leave the Insti­

tute next year has caused this review to contain various articles with a final char­

acter, viz. stereotyped responses and mental load. In other studies the effort is

aimed at better and more sophisticated tools, as is the case with handwriting, tremor,

electromyography (EMG) and the measurement of muscle characteristics. In the ergono­

mics field the tendency has persisted to combine the available manpower in only a

few projects. The main effort has been made in the videophone project and in the

medical field.

stereotyped responses

In previous Progress Reports (~, 1970; ~, 1972) the development of stereotyped res­

ponses has been described, to be understood as overlearned responses of well-trained

subjects.

Most experiments involved the coding of 4 letters with digits, according to two codes,

one being the reverse of the other. Coding the letters from lists that were scanned

as in reading, showed that after several thousand trials code responding approached

the speed of normal reading. This was the case for both coding schemes, which were

alternated session for session.

In another experiment letters were presented individually and reaction time measure­

ments were taken. Only one code was learned, followed by the other after 5500 trials.

Reaction times in the latter case remained high after 1600 trials; however, in both

codes the error rates decreased clearly. From this it may be inferred that a response,

once stereotyped, interferes with a conflicting non-stereotyped response, but need

not interfere with its correctness.

To check this hypothesis, an experiment was carried out in which the two conditions

were combined:

(I) overlearning one code, then switching to the other,

(II)learning the two different codes alternately, both conditions in the reaction

time paradigm.

The results will be published elsewhere.

handwriting

Pulsation

From earlier studies on handwriting (V.redenbregt and Koster, I.P.O. Annual Progress

Report ~, 1967) there was evidence that during handwriting the electromyogram (EMG) ,

is of a pulse-like character. The amplitude of the pulses appeared to be relatively

constant, whereas pulse duration varied. This phenomenon, called pulse-width modu­

lation, has also been the basis of a handwriting simulator. In order to test whether

such a denomination is a realistic one, a pilot experiment has been carried out con­

cerning a very simple movement, viz. flexion and extension of the lower arm, the

amplitude and the period being varied. The movement itself and the EMG were measured.

In the case of very long periods, viz. about 1s,hardly any modulation of the EMG was

noticed. With decreasing periods the modulation increased and from about SOO ms,

clear pulsation ~as observeu. The duration and the mean amplitude of the EMG pulse

IPO annual progress report 8 1973

43

was measured. Independently of the amplitude of the movement, the shortest period of

movement at which pulsation was observed was about 180 ms. As there is apparently

a minimum duration of about 100 ms for the pulses, faster movements are not possible.

In fact, in that case the pulses of agonist and antagonist start to overlap and thus

counteract. Fig. 1 shows the EMG amplitude and the EMG duration in relation to the

period of movement.

DURATION EMG[ms J

AMPLITUDE EMG[arbitrary units]

600

400

200

oo 200 400

PERIODo

OF MOVEMENT200[m s]

400

Fig. 1. Duration and amplitude of EMG pulses during flexion and extension of the arm.

Durational build-up

In I.P.O. Annual Progress Report ~, 1967, a study was described in which were measured

the intervals defined by the points of time at which the writing movements reversed

their direction. The ratio between these intervals appeared to be approximately con­

stant for all five subjects. The same constant ratio was found irrespective of writ­

ing speed and size. A model of the durational build-up of these intervals has been

developed with the help of Mr. H.N. Linssen of the Dept. of Statistics of the

Eindhoven University of Technology.

In order to test its predictions for large variations in writing time, a pilot study

was carried out in which character size and writing speed were varied.

Character size was indicated by parallel lines with 3, 8 and 13 mm separation and the

speed instruction was worded by "low", "normal" and "high". Subjects got the sizes

fairly well, but not the writing speeds.

In each speed condition (low, normal and high) the writing duration was found to be

the same for all three character sizes. A larger character size written with the

same speed as a smaller one, should, however, result in a longer duration. Subjects

apparently misinterpreted the speed instruction.

The data mentioned above have been analysed from the recorded displacement-time dia­

grams by means of a ruler. In order to avoid such time-consuming activities, a com-

44

puter programme has been written for automatic analysis of handwriting data. A new

device for picking up handwriting movements has also been developed by Mr. C. Huber

of the Department of Electronics of the Eindhoven University. A small solenoid is

moved in a very stable electromagnetic field whose strength depends on the distance

from a reference line.

writing errors

In a previous issue attention was focussed on the kinds of errors that occur during

handwriting (Van Nes, 1971). Recently the position of errors in the text was studied.

For example, frequencies of error-occurrence was determined in relation to letter

position within a word. It has been shown that up to a certain wordlength, the cen­

tral letters are more prone to erroneous writing than the first and last letters.

Letters in the middle of long words, however, may show the same low frequency of

errors as first and last letters. Such data are interesting in view of the span of

interaction between representations of the letters of a word in the motor programmeconcerned.

keying errors

Besides writing errors a study has been undertaken on keying errors made by card

punch operators. The aim of the study was to reduce the number of errors through

knowledge of the error sources. About 65,000 keystrokes made by four operators were

analysed. A total of 292 errors were made, of which 222 were detected by the opera­

tors themselves. Two main categories were established, viz.

(a) symbol sequence errors (38%) and

(b) errors arising from various types of mis-reading the documents.

Mis-reading single symbols only accounted for about 3% of all errors. Inaccurately

aiming at the proper key, i.e. hitting a neighbour, caused 11% of the errors. Thepercentage of non-detected errors was about the same for both categories. Hence,

most of these errors were the result of mis-reading. Better readable documents are

expected to decrease such errors.

mental load

In previous issues of the Progress Report (Koster, 1966; Koster and Ivens, 1968)

studies have been published about the measurement of mental load, and in particular

of heart rate variability (sinus arrhythmia) as a possible indicator of mental load.

A general finding was that after training no clear effect of changes in task load

was reffected in the variablity of heart rate. With respect to the rest condition,

often a decrease in heart rate variability was observed. A classical task in such

studies is the binary choice task.

According to a pre-determined time system as Mentat Work Factor this task requires only

little time, whereas discrimination of small changes between stimuli (for instance,

in the diameter of a spot) requires much more time. Discrimination as an experimental

variable has the advantage over the binary choice task that the time needed for the

task is mote dependent on mental factors than on physical factors.

Therefore an experiment was carried out in which the diameter of a circle was random­

ly varied around a mean value. The discrimination time and the error score were

measured at each presentation. During the task proper the beat-to-beat values of the

heart-rate were measured as well. In all conditions a standard circle of 9 mm (view­

ing distance 3.10 m) was presented. The test circle appeared at a distance of 0.20 m

4S

below that circle. The diameter of the test circle was variable from trial to trial

and chosen from a normal distribution with 9 classes. Four levels of "load" were

introduced by varying the standard deviation of the diameter of the test circle, viz.

O. 6; 1. 2; 1. 8 and 2.4 mm.

Two highly trained subjects participated in the experiment. Series of 100 trials each

were performed, arranged according to a Latin Square design. Each "load" condition

lasted about S minutes, it was preceded and followed by a rest period of about 3 min­utes.

With increasing "load", i.e. with decreasing standard deviation of the diameter,

error scores and RT increased clearly. No such effect could be established with sever­

al sinusarrhythmia scores. The scores have been tested both absolutely as well as

relatively to the value of the preceding rest. With respect to the preceding rest

trials in nearly all "load" trials a suppression of heart-rate variability of about

30% was measured. This suppression appeared to be independent of the "load".

As a preliminary conclusion it may be stated that even with this task, which was

thought to be a difficult one, no appreciable effect on sinusarrhythmia was establish­

ed by varying the "load".

After this experiment we are even more sceptical about the value of sinusarrhythmiaas a measure of mental load (whatever that may be).

tremor

In the 1972 issue of the Progress Report two articles were published on tremor

(Koster and Van Schuur, 1972 a, b) both describing the results of experiments in

which a limb was loaded with different weights. The data show evidence that physio­

logical tremor is nothing but the mechanical response to a random forcing signal,

whereas in the case of Parkinsonism a "nervous" hypothesis was presented. EMG as

well as ballistocardiac activity can serve as a forcing signal. To discriminate be­

tween these two, a more subtle experiment had to be performed.

A disadvantage of the previous experiments was that by placing weights on the index

not only the dynamic load (inertia) was increased but also the static load (mass).

To disentangle effects of mass and inertia, a new set-up was chosen in which the

weight of the finger could be balanced to avoid unnecessary muscle activity. From

some pilot experiments with a balanced finger it was clear that the amplitude of the

mechanical tremor is very small indeed. As a result, the acceleration signal becomes

very small and the signal to noise ratio drops to such a low value that it was de­

cided to measure the displacement of the finger instead.

A special device has been built which is shown in Fig. 2.

The basis is a commercially available displacement measuring device (Pretec) capable

of measuring displacements to at least 100 Hz. The finger is connected to the beam

of a very light balance, whose deviations are measured. Through an adjustable mass

the weight of the finger can be balanced. Two other weights can be used to load the

finger with an amount of inertia. This set-up allows of measuring movements of the

finger from 1 ~m up. The equivalent mass of the device in the balanced condition is

about 30 g.

From coherence measurements of the data published in Annual Progress Report 7 (1972),

it became apparent that the coherence between EMG and tremor acceleration was about

twice that between the envelope of the EMG and tremor acceleration. This means that

the EMG signal itself and not the envelope must be considered as the input signal.

46

Fig. 2. Device to measure very slight movements of the finger.

When the finger is loaded, tremor signals of about 5 Hz are observed. Together with

the small EMG values in the unloaded, balanced condition this puts rather strong

requirements on the EMG pre-amplifier. A third aspect of this amplifier that might

be of importance is the input impedance. In a pilot study two identical amplifiers

were used, one with an input impedance of 1 M ~, the other with 200 M ~. Using

Ag/AgCl electrodes, no differences were detected, neither in the recording nor in

the power-spectrum. Using conducting rubber electrodes, the data of both subjects

showed low frequency « 10 HZ) disturbances every now and then. For clinical pur­

poses this can hardly be of importance assuming the amplifier cuts off at 15 Hz. For

our purposes we preferred the 200 M ~ input impedance.

Keeping all these requirements in mind, a new amplifier was developed with an input

impedance of 200 M ~, a noise level ot 10 ~V and a frequency characteristicr .m. s.flat from 1 - 750 Hz within 3 dB. The instrumentation that is now made available

makes possible a thorough study of slight and low-frequency tremor.

muscle studies

In Annual Progress Report ~ Vredenbregt and Koster (1966) described a device for

measuring simultaneously the degree and rate of contraction, the acceleration and

the force at the wrist during flexion and extension of the forearm. The device has

been used extensively in the last few years. It has, however, some limitations as a

result of higher requirements made. An important point was the reduction of inertia,

which was 1.8 kg equivalent mass at the wrist. A reduction of the friction

47

in the bearings was also achieved. The employment of various subjects revealed an­

other restriction, viz. the beam of the device could not be adjusted for subjects

with different armlengths.

These requirements were met by using the light-weight and very strong tube construct­

ions developed for space-flight purposes together with high-quality ball bearings.

At the same time the device was made portable, which is often desirable when patients

serve as subjects. The result is depicted in Fig. 3.

Fig. 3. Portable device for measuring characteristic of elbow flexors and extensors.

The upper arm rests on a support that is adjustable in height. It is curved at the

arm-pit for better transfer of large forces. The distance between the axis of rotat­

ion and the cuff as well as its parallel beam at the rear side of the device, is ad­

justable to various arm sizes.

The pick-ups, i.e. position sensor, speedometer, accelerometer and dynamometer, have

been reduced in weight and their sensitivity has been increased.

ergonomics

The available manpower has mainly been concentrated upon two projects, viz. the vi­

deophone and a medical treatment system.

Videophone

The last two years the Institute has been involved in the set-up of a videophone trial

network, a joint venture of Philips and the Netherlands P.T.T. administration. The

network will connect 50 subscribers in 5 different locations. In 1970 and 1971 the

videophone set proper was developed. Ergonomic considerations aimed at an optimal

viewing distance from the screen and tests were performed with different loudspeaker

positions. A number of additional facilities, such as Zoom Focus, Document Display

and Selfview were designed to make operation easy.

48

An important aim of the trial network is the study of subscribers' reactions to the

new dimension in station communication. Since 1970 an evaluation team of Philips and

P.T.T. members have been preparing this study. The first step was the selection of the

various videophone stations within the area covered by the network. Starting from

the number of telephone calls between departments of the two organisations and from

the number of meetings between individuals, an optimal user population was established.

It was also considered useful to connect departments of diverging character such as

technical, commercial, administrative and medical.

The evaluation team is preparing the instruction and follow-up of participants, traf­

fic observation, interviews, studies on user behaviour and communication patterns,

and laboratory experiments. The network will be in operation as from spring 1974 andwill remain in for two years.

Medical tpeatment system

At the request of the Philips Medical Systems Division the use of a system for radiat­

ion treatments has been studied in two hospitals. The object of the study was:

(a) to investigate how existing systems are used and how staff (therapeutist, physi-

cist and nurses) and patients behave during a treatment,

(b) to gather opinions of staff and patients about the system,

(c) to collect information on possibilities of automation of operating mechanisms.

The study has been carried out by means of video recording and through interviews.

The use of video recording has several advantages. Two fixed cameras were used, one

with a wide-angle lens, the other with a telelens. This set-up appeared hardly to

disturb the relation between doctor and patient. Both preferred this set-up over the

bodily presence of an observer. Video recording presents the possibility of observing

the same scene repeatedly with different ends in view. It was found valuable to com­

pare situations undisturbedly. In all pictures the patient's face was made unrecog­

nisable.

The video tapes also proved useful to convince the designer of the benefit of pro­

posed modifications. The system designers obtained a better sense of the use of their

system.

It appeared that both staff and patients were often eager to give comments on the

system and its use.

Video recording and interviews have proved to produce valuable information in view

of new developments.

5 instrumentation

49

INTRODUCTION I.P.O. INSTRUMENTATION 1973D.J.H. Admiraal

After the arrival in 1970 of the computer there was an increasiijg need to use this

machine - apart from actual calculating operations - for on-line experiments. This

has led to a continuous reconstruction and adaptation of the machine in such a way

that it is no longer necessary to carry out experiments in the computer room itself

but in any arbitrary room in the laboratory. This possibility requires a number

of interfaces for sending and receiving all sorts of data.

The transmitter and receiver for such data in the computer itself was realised in

1971 (I.P.O. Annual Progress Report 6, 1971). The data transmitter and receiver in

the experimenter's room has been called MARIE, a Dutch abbreviation of the English

equivalent: Modular Adapter between Computer Interface and an Experiment. MARIE is

now with various modules and a control desk ready for experiments. The nature of

each experiment defines what modules are necessary. For acoustic research are now

available:

a six-channel electronic ON/OFF switch with FET's,

a digitally controllable attenuator of 63 dB in steps of 1 dB,

a module which can control the Frequency and the Amplitude of a Frequency

Synthesiser, and

some other modules.

All these modules belong to the Philips 19" - 3E mechanical building system, which

offers great flexibility and a simple way of exchanging units. This modular system

was used for the first time at the I.P.O. in the Digital Timer (I.P.O. Annual

Progress Report l, 1968).

In this chapter a description is given of an Analogous Timer Unit CIDAT, a Dutch

abbreviation of the English equivalent: Continuously Adjustable and Digitally

Readable Timer. After receiving a starting pulse, this module generates an output

pulse after t seconds. The delay time t is adjustable very accurately by means of a

10-turn potentiometer and can be read directly in 3 decades on Numerical GaAs

Indicators. As applications in research can be mentioned:

the temporal adjustment of two stimuli,

the location of a segment in a speech utterance, and

the adjustment of the duration of gated speech signals.

As the counter can be used separately, the display makes possible its use as a trial

counter in an experiment or as a reaction time measurement device.

In experiments where signal frequencies have to be adjusted the digital display can

be used to advantage with respect to pointer reading.

Last year much attention was paid to the recording of Eye Movement (See: Introduction

Visual Research, elsewhere in this issue). The work covers two techniques:

(a) a detection of the position co-ordinates of the so-called Cornea Reflection spot,

recorded by a Silicon TV-camera,

(b) a method based on the Limbus reflection, especially in order to see whether it is

possible to arrive at a recording device which can be carried by a subject

in the form of a pair of spectacles or a kind of helmet, so as to avoid the un­

pleasant use of a head and/or chin rest and/or a bite-board to stabilise the

position of the subject's head.

So far we used the well known Cornea Reflection method. Briefly, this technique

/PO annual progress report 8 1973

50

supplies an electronically mixed signal of two TV cameras, one for the corneal

reflection and one focussed on the scene. It gives TV pictures in which the point the

subject fixates is represented by a bright spot super-imposed on the image of the

scene. The TV images are recorded on video tape. Then the positions of the spot were

determined by slow-motion play-back of the video tape, a monotonous and time con­

suming job.

Today the x and y ordinates are determined electronically to find the positions of

the spot. Interface circuits for feeding the computer are under design.

The technique of the Limbus reflection offers more problems, especially at the optical

side. Two infrared light strips are projected on the Limbus (lower Iris side). The

amount of reflection is measured using two very sensitive photodiode amplifiers. As

chopped light emitted by power LED's is used, the first part of the amplifiers can be

AC-coupled, owing to which the influence of room straylight is suppressed.

After synchronous detection and lowpass filtering two DC voltages are available which,

after addition and subtraction are a measure of the y and x ordinate voltages.

At the optical side split-fibres are used in which input and output wires are mixed

in the slit-shaped end. Should it appear that the reliability of this system is

sufficient, it will also be coupled on-line to the computer.

51

CIDAT, A CONTINUOUSLY ADJUSTABLE TIMER WITH DIGITALDISPLAYD.J.H. Admiraal

In research work, Timers are necessary to:

(a) generate a time period, e.g. the duration of a stimulus,

(b) generate a delayed point of time, e.g. the stimulus appears a number of ms after

a reference point of time.

Timers can be:

(a) digital, which means that the generated time durations are a multiple of a given

time interval, derived from a crystal clock,

(b) analogous, which means that the generated time durations can be varied continuous-

ly within certain limits. This is realised in the Timer CIDAT described below.

The digital Timer in a modular form, developed at the I.P.O. (Valbracht,I.P.O. Annual

Progress Report, 1968) consists of a cabinet containing a supply and a crystal clock

module, together with some counting modules. With one such counting module it is

possible - by adjusting the two thumbwheel switches and by choosing one of the avail­

able time intervals 0.1 - 1 and 10 ms generated by the clock - to obtain time inter­

vals between 0.1 and 999 ms. The intervals are very accurate, stable and easily re­

producible. The great practical advantage of this modular system is that other

modules than those mentioned above can be simply shifted into the cabinet, using in

this manner the already available supplY,clock and/or counting module(s).

In I.P.O. Annual Progress Reports the following modules for this system have beendescribed:

1969: A Modular Exponential Function Generator

1970: POBU, a Pulse on Tape Unit

The Vario-S-Gate

1971: The LED-Gate for Audio Signals

The FET-Switch and some of its Applications

A Trapezium Function Generator

1972: A Pulse-Width Light Modulator

A new module in this range is the Analogous Timer eIDAT. In response to a starting

command CIDAT generates a final pulse after a certain interval t which can be readon numerical indicators (See Fig. 1).

Starting command = reference point of time

t -, Time interval t, to be adjusted by a potentiometer

o

/PO annual progress report 8 /973

Final pulse of Timer • delayed starting pulse

Gate for clock pulses

Clock pulses passing through the gate

Counter with numerical indicators.Only the final value of the counter is made visible.

S2

with the thumbwheels.and faster than is possibleC LOC K INTERVALS

DIGITAL TIMER

Compared with the Digital Timer, CIDAT offers the following advantages:

(a) a separation is introduced between the adjustment - with a knob - of the generated

time intervals and the display. As a remote control of the timer is possible, the

adjustment can be effected by a subject while the experimenter observes the ad­

justed time intervals,

(b) the adjustment is much easier

Fig. 2 gives the block diagram of CIDAT.

0.1 10ms DISPLAY

GE

SINE/PULSE -.II J 4 1 7r"V- l-- ~

I CONVERTOREXT.

II I Iu- Sl

START-.GATE FLlp·FLOP

GATE AND}--STOp-r-- F

COUNTER of 1,000FINAL--PULSE

P ~~

ONE-SHOT

~I~ ,. ~ - TIMER -.......12 fJS,

LATCHADJUSTMENT VOLTA

f--- -____S.1 _________

START • • • 1F.C. C.P.-TIMER TFIG 2

The starting pulse is supplied to a Timer generating a delay time t which can be

accurately adjusted with a 10-turn potentiometer P over a range of 1 to 100. At the

end of t the Timer resets automatically and triggers a one-shot which generates

the Final Pulse, which is externally available. The starting pulse can simultaneous­

ly be supplied to a bistable flipflop F. In the 1-state of F an AND-gate is opened,

passing through clock pulses from the digital timer to a counter of 1000.

The digital output of this counter, in BCD-code, is supplied to the Display with

Hewlett-Packhard Solid State Numerical Indicators. Each indicator - in 0.4" x 0.6"

dual-in-line envelope - is provided with an internal Decoder to convert the binary

code at the input to the decimal code .for the 4 x 7 matrix of Gallium-Arsenide

Light Emitting Diodes (LED's).

Besides, each indicator has an internal memory with latch. When the latch voltage is

low « + 0.8 V), then the memory register follows the information at the input

"real time". If the latch voltage becomes high (> + 2 V) then the information from

the input to the memory is stopped and the position of the counter is displayed as

it was just before the moment the latch voltage became high.

At the end of the delay time t the one-shot of 12 ~s resets the gate flipflop so that

the gate closes. During the time that this one-shot is set the latch voltage is kept

low. This means that only the final value of the counter is made visible. Only

changes in final values are displayed, stimulating quiet reading.

S3

The decimal points on the numerical indicators are used to indicate when the gate is

open. As applications can be mentioned the temporal adjustment of two stimuli, the

localisation of a segment in a speech utterance, and the adjustment of the duration

of gated speech signals.

other possibilities of cidat

Time-interval measurement

Switch S2 in Fig. 2 at the position CP (counting of pulses). In this position the

latch voltage is kept low, resulting in a real time display of all counting positions.

At the same time the Start-input to the Timer is interrupted so that no final pulse

can appear. At the moment of the starting pulse, which can initiate for instance the

stimulus, the gate flipflop F sets and opens the gate. This closes at the moment

that F resets at a stop pulse at the corresponding bus as a reaction by a subject.

In this manner reaction times can be measured and displayed directly.

The fact that in addition to the final value all other positions of the counter are

also displayed is of no great importance to this application.

Counter of pulses

As above, but now with switch 52 in the position EXT. After a starting pulse the gate

opens and the counter together with its display now follows pulses real time supplied

to the bus marked EXT. Under these conditions eIDAT measures for instance the number

of times a subject has responded to stimuli in an experiment.

Gate and Timer in series

This condition is possible because both have separate inputs. An example of applicat­

ion is an experiment in which a subject operates in an self-paced condition. At a

moment which he decides himself he presses a knob through which the Timer starts.

After a delay time t, to be varied continuously by the experimenter, a final pulse

is generated, initiating the stimulus and setting flipflop F with gate. The latter

closes by a reaction pUlse given by the subject. According to this procedure it is

completely impossible for the subject to synchronise on the starting pulse.

Frequency Counter

CIDAT can also be used as a Frequency Counter. In this case suitable output pulses

of a digital Timer are used as starting pulses for one counting interval.

During the 12-~s starting pulse the latch voltage is made low so that the result of

the previous counting is made visible, while the counter is reset. The signal sourcethe frequency of which is to be measured is connected to the input marked EXT. In

this case the frequency of a signal source can be measured directly in three decades

without consulting the often inaccurate scale of the source itself.

references

Valbracht, J.e. (1968) MTG - A Modular Time Source, I.P.O. Annual Progress Report.~, 113 - 114.

S4

MARIEXINTERFACE BETWEEN COMPUTER Al\ID EXPERIMENT

G.J.J. Moonen and Th.A. de Jong

END OFEXPERIMENT

Fig. 1. Flow diagram

In the perception research many experiments have become so complicated that control

by a computer has become a necessity. In fact, this is the case with the adaptive stra­

tegies for experiments, for instance for the determination of thresholds. The computer

can thus handle the generation of the stimulus (e.g. tone bursts or light flashes)

as well as the digestion of the responses (i.e. the subject has or has not perceived

the stimulus). Depending on the response of the subject, the computer determines the

threshold by increasing or decreasing the level of the next stimulus according to an

adaptive method (see flow diagram Fig. 1). In addition the computer can decide very

rapidly so that the time required by each trial is

very short, in consequence of which the attention

of the subject will not slacken. Therefore, MARIE

has been designed. The purpose of MARIE is to control

measuring equipment outside the computer room to ob­

tain better experimental conditions. The information

transport between the computer and MARIE is realised

digitally. In the experimentation room MARIE passes

the information on to the equipment, and vice versa.

The information of the computer may control the digital

instruments of the equipment (e.g. a frequency syn­

thesiser or an attenuation module) by means of

adapting modules. In the reverse direction these

EXPERIMENTATION ROOM

MEASURINGIEQUIPMENTII

I max~ 100m ICOMPUTER~~~--~IMARIE

Fig. 2. Survey of the equipment

modules convert the output signals from

the experiment into digital information,

which passes from MARIE to the computer

(e.g. an analogue-digital converter).

Each module is selectively addressable

by the computer. To keep the dimensions

of MARIE limited, we decided on a modu­

lar way of connection between MARIE and

the experimental equipment (up to 8

modules). The experimenter himself can make a choice from the available modules and

can also use modules having a particular function. The following modules are

available:

Analogue-digital converter (e.g. for feeding speech signals into the computer)

Digital-analogue converter (for producing speech signals)

Programmable timing unit (for time intervals or duration of stimuli)

Programmable attenuator (attenuates electric signals)

Digital controlled FET switches (to make a choice out of 6 signals)

Adapting modules are available for:

Frequency synthesiser

Frequency counter

12-channel digital input (e.g. push buttons)

12-channel digital output (e.g. lamps or relays)

x MARIE is the Dutch abbreviation of Modular Adapter between a computer and anexperiment.

IPQ annual progress report 8 /973

Here the central dis­

tributing device ofMARIE ~ith severalmodules and the man­ual control panel are

sho~n.

55

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

56

general description of the system

EXPERIMENT

16

16 bl. t

in

output

COMPUTER MARIE

send the module address and after

that the information of that mo­

dule, or to have the computer as­

semble the address and the infor-

mation in one word. MARIE then

Fig. 4. Block diagram separates the assembled information

into an address part and an infor­

mation part. The latter solution is accepted because in this way we can transfer the

information almost twice as fast as in the other case. A disadvantage, however, is a

decrease in information accuracy. A reasonable compromise between the number of

addressable modules and the accuracy of information is found in dividing the word

information into a part of 4 bits for addressing and a part of 12 bits for information.

In this way a 0.25 % inaccuracy can be obtained. If a still lower figure is required

several modules may be combined.

Thus the computer can supply output directly because both the address and the in­formation are assembled in one word. More complicated is the process of asking infor­

mation of a module. In this case MARIE has to be informed about which module is askedfor before the information can be taken in by the computer. Therefore, the address of

the module must be announced to MARIE via the output channel. The information plus

the address of that module is then sent to the computer.

A provision has been made for putting a module which is to be read out several times

successively, in the "repeat mode". Then this module has to be addressed only once.

In addition to three separate addressable registers, which provide the possibilityof interfering with the experiment from a manual control panel, MARIE has been fitted

with a status-register, which is to be read out by the computer as desired. It indi­

cates that the modules are ready for loading or reading.

Fig. 4 depicts the block diagram of the system. The computer has a wordlength of

16 bits. The information transport from and to MARIE is effected along twice 16

twisted pairs of wires excluding some interlocking wires.

To transfer information to a speci­

fic module, the address of the

module must be given to MARIE. Two

obvious solutions are: First to

6 ipo publications 1973

57

I.P.O. PUBLICATIONS 1973P 231 J. Vredenbregt, G. Rau

Surface Electromyography in Relation to Force, Muscle Length and Endurance.

J.E. Desmedt (Ed.) New Developments in Electromyography and ClinicalNeurophysiology, Vol. 1, Basel, Karger, 1973, 607-622.

During static (isometric) contractions of the flexor muscles of the forearmthe force was measured at the wrist simultaneously with the EMG activity fromm.biceps and brachialis.When the electrical activity of the biceps was related to the ratio betweenthe force exerted at various levels of activity and the maximum force at thesame muscle length, all curves appeared to coincide without any change in thestandard deviation.During voluntary static contraction at constant force exerted as long aspossible, an almost linearly increasing EMG activity was observed whichdepended on the force sustained.If the EMG is kept constant, the force as a function of time decreases in acharacteristic manner. The decrease was measured at different values ofconstant EMG activity.These results suggest that under static conditions an unequivocal relationbetween EMG and force exists. This suggestion is strengthened by the resultsfrom the experiments in which the constant force was changed suddenly fromone level to another. The increase in EMG activity at the two levels ofactivity was in full agreement with the results obtained during sustainedconstant force, and depended on the momentary force.

P 232 S.G. Nooteboom

Some Timing Factors in the Production and Perception of Vowels.

Occasional Papers, University of Essex, 1972, nr. 13, 49-76.

A related publication can be found in I.P.O. Annual Progress Report ~, 1972,30-39.

P 233 I.H. Slis

The Influence of Articulatory Effort on the Timing of Speech.

Occasional Papers, University of Essex, 1972, nr. 13, 128-150.

A related publication can be found in I.P.O. Annual Progress Report ~, 1972,40-44.

P 234 J.A.J. Roufs

Dynamic Properties of Vision - III. Twin Flashes, Single Flashes and Flicker­fusion.

Vision Research, 1973, ~, 309-323.

The threshold of two identical rectangular incremental flashes or "twinflashes" was measured as a function of the interval between the flashes. Themeasurements were effected with a 1° foveal stimulus for various backgroundintensities. The relationship with threshold characteristics for singleflashes of variable duration, measured by the same subjects under identicalconditions, is demonstrated. The characteristics accord with the theoreticalvalues calculated on the basis of the theoretical model discussed in Part II,from the subjects flicker-fusion curves determined under identical experimen­tal conditions.

P 235 F.F. Leopold, H.J. Bleileven

Human Factors in the Development of an Emergency Telephone System.

Sixth International Symp. on Human Factors in Telecommunication, June 26-30,1972, VI.2, 1-5, Stockholm

P 236 S.G. Nooteboom

The Perceptual Reality of Some Prosodic Durations.

Journal of Phonetics, 1973, l, nr. 1, 25-45.

Some questions are raised pertaining to the internal representation ofsyllable nucleus durations, being part of words produced in isolation. Afirst question concerns the accuracy with which a syllable nucleus durationmay be internally represented. Further questions concern some major prosodicregularities which were discovered in earlier articulatory measurements onDutch nonsense words. It is asked whether these regularities are part of the

IPO annual progress report 8 1973

58

internal representation language users have of how words in their languageshould sound. These questions are studied with a method in which subjects areasked to adjust the duration of one syllable nucleus in a synthesized wordaccording to some internal criterion. It is found that the internal represen­tation of a syllable nucleus duration may be more accurate then the spectro­graphic measurement of its acoustic correlate. The internal representation ofhow words shouls sound appears to be governed by rather strict temporal patterns,of which phonological vowel quantity, stress and position in the word are im­portant determinants.

P 237 R. Collier, J. I t Hart

Perceptual Experiments on Dutch Intonation.

A. Rigault and R. Charbonneau (Eds.), Proceedings of the Seventh Int.Congress of Phonetic Sciences, University of Montreal, 22-28 August 1971,The Hague, Mouton, 1972, 880-884.

P 238 J. 't Hart

Intonational Rhyme.

Acta Univ. Carolinae, Phil. 1, Phonetica Pragensia III, 1972, 105-109.

Paper presented at the Symposium on Intonology, Prague 6-8 Oct. 1970.P 239 A. van Katwijk

On the Perception of Stress.

Acta Universitatis Carolinae, Philologica 1, Phonetica Pragensia III, 1972,127-135.

Paper presented at the Symposium on Intonology, Prague 6-8 Oct. 1970.P 240 H. Bouma

Visual Interference in the Parafoveal Recognition of Initial and FinalLetters of Words.

Vision Research, 1973, ~, 767-782.

As part of a visual investigation of reading processes, single words havebeen recognized in parafoveal vision. Correct initial and final letters arecounted in the response words. Surprisingly, the most outward letters, farfrom fixation, score higher than the most inward letters, close to fixation.The same result is obtained if these letters are directly recognized fromunpronounceable letter strings. This reveals a strong adverse interactionmainly acting towards the fovea.Words and inward letters are recognized better Right than Left of fixation,in particular for long stimuli. Therefore the range of interaction seemssmaller Right than Left of fixation possibly due to reading habits. Lettercounts are higher in words than in letter strings. The difference gives thecompletion by other perceived word properties.

P 241 W.G. Koster, R. van Schuur

The Influence of the Intensity of Tone Bursts on the PsychologicalRefractory Period.

S. Kornblum (Ed.) Attention and Performance IV, New York, Academic Press,1973, 55-69.

An experiment is described in which the subject must react to the second oftwo similar stimuli (1000 Hz, 15 msec). The intensity of the stimuli and theinterval between them are randomly varied. The results are in contradictionwith the intermittency hypothesis. They may be explained in terms of asensitivity hypothesis. The decay in reaction time as a function of theinterstimulus interval depends on the experimental conditions which suggeststhat other influences also playa part in such experiments.

P 242 S.G. Nooteboom

Temporal Patterns in Dutch.

Proceedings of the Seventh Int. Congress of Phonetic Sciences, held at theUniversity of Montreal 22-28 August 1971, The Hague, Mouton, 1972,984-989.

P 243 S.G. Nooteboom, I.H. SlisThe Phonetic Feature of Vowel Length in Dutch.

Language and Speech, 1972, ~, 301-316.

/PO annual progress report 8 /973

S9

A basic assumption of the research reported upon here is that measurable. vowel duration is at least partly controlled by an independent phoneticfeature of vowel length. We have studied manifestations of this feature bymeasuring articulatory segment durations of a number of Dutch short vowels,long vowels and diphthongs, in nonsense words of the form /pVpVpVp/ of whichthe second syllable was stressed. /V/ was the same vowel in each syllable.The /p/ durations and the total word durations were also measured. Thedurations were defined by the moments of lip opening and closure measuredwith a lip contact. The results can be explained by assuming that on a pro­gramming level of speech production there are only two possible specificationvalues of the feature of vowel length for Dutch vowels, one for the shortvowels /a,';) ,re, I, e:, u, y, i/ and one for the long vowels /a:, 0:,1'1:, e:/,plus the diphthongs /au, Ay, e:i/. The programming level of speech productionis taken to correspond to the linguistic level of phonetic representation.Durational variations within each of the two phonetic categories of vowellength can be explained by peripheral properties of the speech productionsystem. Variations due to stress and position are accounted for by assumingthat prosodic patterns may affect the amount of d~viation of the actualduration from the specified ideal duration.

P 244 H. Bouma

Der Einfluss zweier Mydriatica auf die Statischen Lichtreaktionen derMenschlichen Pupille.

Symposion der Deutschen Ophthalmologischen Gesellschaft, Bad Nauheim,Munchen, Verlag J.F. Bergmann, 1973, 216-221.

P 24S J. Vredenbregt, G. Rau

Muscle Coordination in Simple Movements.

Third Intern. Seminar on Biomechanics, Rome, September 1971.E. Jokl (Ed.) Medicine and Sport, vol. 8: Biomechanics III, Basel, Karger,1973,239-242.

P 246 G. Rau, J. Vredenbregt

EMG - Force Relationship during Voluntary Static Contractions (M. Biceps).

Third Intern. Seminar on Biomechanics, Rome, September 1971.E. Jokl (Ed.) Medicine and Sport, vol. 8: Biomechanics III, Basel, Karger,1973, 270-274.

P 247 J. 't Hart, A. Cohen

Intonation by Rule: a Perceptual Quest.

Journal of Phonetics, 1973, l, 309-327.

An approach towards the study of Dutch intonation which has been called"perceptual analysis" has revealed the regular occurence of a rather simpleconfiguration, resembling a hat-shape. This paper presents a number of rulesaccording to which pitch contours for arbitrary utterances can be derivedfrom this pattern. When tested on several kinds of new material, these rulesturn out to be too rigid.On the same basis, other equally acceptable contours may be formed. In thematerial used, these alternative shapes do not give rise to changes in inter­pretation. On these grounds, the rules have been adjusted and extended with anumber of "optional alternative rules".

doctoral thesis

J.A.J. Roufs

Dynamic Properties of Human Vision.

University of Technology, Eindhoven, November 1973.

/PO annual progress report 8 /973

61

reprints of i.p.o. publications

Requests for reprints of publications listed on pages 58-60 can be stated in the

form appearing below by mentioning the publication number.

Reprints of the contributions appearing in this volume may also be ordered.

Back numbers and reprints may be obtained from:

LibraryInstitute for Perception ResearchInsulindelaan 2EINDHOVEN 4502The Netherlands

type-print or capitals please

NAME

FUNCTION and/orINSTITUTE

ADDRESS

CITY

(STATE/ AREA CODE) .....•...........•.......•...

COUNTRY

would appreciate receiving a reprint of the following publications:

numbers (or title and author(s)):

Signature: