IMPROVING SPELLING SKILLS FOR BLIND

LANGUAGE LEARNERS

∗

Orthographic Feedback in an Auditory Vocabulary Trainer

Verena Stein, Robert Neßelrath and Jan Alexandersson

German Research Center for Artiﬁcial Intelligence - DFKI GmbH, Saarbr

ucken, Germany

Keywords:

Vocabulary acquisition, Orthography, Computer assisted vocabulary learning, Visual impairment, Auditory

feedback, User centered design.

Abstract:

This paper describes the design and the design process of orthographic feedback in a computer-assisted vo-

cabulary learning (CAVL) application that is targeted at blind language learners. It discusses current research

ﬁndings of vocabulary and spelling acquisition, as well as special needs of blind computer users. CAVL appli-

cations often assume the user’s sightedness. While it is possible for blind users to access software via screen

reader or Braille line, it is argued that orthographic feedback does not translate one-to-one from visual-to-

auditory (or tactile). To overcome this short-coming and thus ensure high usability for blind users, the feed-

back may be designed differently to be delivered via the auditory channel. Following a user-centered design

approach, the orthographic feedback in an auditory vocabulary trainer is constructed and evaluated iteratively

with users. The preliminary architecture that evolved out of these pre-studies is reported and discussed. In

conclusion, an outlook is given as to how the ﬁnal architecture will be implemented and evaluated.

1 INTRODUCTION

Blind people often use text-to-speech synthesis when

they work on the computer. Using a Braille line to

convert digital texts into embrossed writing is slow

and requires extra hardware. Relying more on TTS

technology, however, lessens blind people’s exposure

to the written form of words which is especially im-

portant when studying another language. The En-

glish language makes it difﬁcult to deduce a word’s

orthography from its pronounciation. Consider ho-

mophones such as ‘seam’ and ‘seem’. Is there a way

for blind people to study meaning and form of vocab-

ulary within an auditory CAVL application?

The aim of this paper is to explore the possibility

of constructing a vocabulary trainer that is both highly

usable and efﬁcient for blind language learners. In

section 2, this endeavor is motivated by providing es-

sential background information in the ﬁelds of vo-

cabulary acquisition, spelling acqusition, computer-

assisted vocabulary learning applications, and special

needs of the blind. The section ends with explic-

∗

This work has been carried out in the scope of the Sen-

sHome project funded by the Saarland Government. The

opinions herein are those of the authors and not necessarily

those of the funding agency.

itly formulating the problem (cf. 2.4). Section3 ex-

plains the methodolody that was used in this study,

one that greatly involves collaboration with the target

users. Section 4 then introduces the conceptional ar-

chitecture and the implementation of the program that

evolved out of these user studies. Finally, section 5

presents the results of the ﬁnal evaluation and section

6 summarises the key ﬁndings of this paper.

2 MOTIVATION

2.1 Vocabulary Acquistion

Building a large vocabulary in a second language (L2)

is essential to language proﬁciency. Especially in

the beginning of learning a new language, every new

sentence contains new words, making it very difﬁ-

cult to deduce the meaning from the context. Fo-

cusing on vocabulary acquisition in this stage is of-

ten a good way to make progress. Advanced learn-

ers also beneﬁt from expanding the size of their vo-

cabulary as it allows them to indulge in the language

at a deeper level. Vocabulary may be acquired both

explicitly and implicitly (Ma and Kelly, 2006). Ex-

501

Stein V., Neßelrath R. and Alexandersson J. (2010).

IMPROVING SPELLING SKILLS FOR BLIND LANGUAGE LEARNERS - Orthographic Feedback in an Auditory Vocabulary Trainer.

In Proceedings of the 2nd International Conference on Computer Supported Education, pages 501-506

DOI: 10.5220/0002860605010506

 SciTePress

plicit vocabulary learning exercises, such as working

with bilingual ﬂashcards, are usually word-centered,

decontexualised and repetitive. Implicit vocabulary

exercises, such as word association games, are of-

ten more complex and provide a meaningful context

for each word. Both learning styles have been ad-

dressed in the design of computer-assisted vocabu-

lary learning tools (Goodfellow, 1995). Computerised

vocabulary study can make learning more interesting

and effective, through means of intelligent interaction

with the user and immediate feedback. While explicit

‘drill exercises’ may fall into the category of “be-

haviourist CALL software” as coined by Warschauer

(Warschauer, 1996), they may actually be more effec-

tive than applications that draw on implicit acqusition

principles. Groot suggests this in the evaluation of a

CAVL application that models the stages of L1 acqui-

sition (Groot, 2000). Simple ﬂash card methods offer

many possibilities when computerised and will be the

primary focus of this work. Cooley points out that

“the users [should be allowed] as much manipulative

freedom with computer mediated ﬂash cards as they

would have with a cardboard version” [p. 3] (Coo-

ley, 2001). One application that implements this idea

without many other gimmicks is Phase 6 (Phase6,

2009). It allows the user to enter their own learn-

ing material and organises the content according to

the “Ebbinghaus curve”, a model of timed repetitions

necessary to retain information in the long-term mem-

ory. The users have to decide themselves whether a

word they entered was correct or not, then the card

is moved accordingly. eSpindle (eSpindle, 2009) is a

more elaborated online application and even though

the learning content is only available in English, “a

strong case can be made for the appropriateness of eS-

pindle as a tool for L2 English learners [. . . ]” [p. 25]

(Olmanson, 2007). What makes eSpindle special is

that actually teaches vocabulary and spelling, by em-

ploying feedback that is instructional as well as mo-

tivational, and by allowing to explore word contexts

via hyperlinks. In this application, it is the system not

the user who decides which answer is correct.

2.2 Spelling Acquisition

Vocabulary knowledge requires knowing both, a

word’s meaning and its spelling. Orthography may be

a tough subject to master, especially when it comes to

“the English language [that] has so many exceptions

to its rules that the rules themselves become mean-

ingless” [p. 18] (Arter and Mason, 1994). There are

differences for those who learn English as a ﬁrst lan-

guage and those who learn it as a second language.

The ﬁrst group will acquire literacy alongside with or-

thography, while the second group will already have

a language that is likely to inﬂuence acquisition of

English spelling to some extent. This work assumes

existing literacy and is concerned primarily with im-

provement of spelling or acquisition of spelling in a

new language. Studies that are concerned with or-

thography acquisition point out the visual nature of

this task (Torbe, 1977), (Arter and Mason, 1994),

(Warda, 2009). Indeed, people form “mental ortho-

graphic images (MOI)” of words (Warda, 2009) and

for those who are not blind or otherwise visually im-

paired, these mental images are likely to be visual rep-

resentations of words. A word can ‘look right’, which

is why people tend to write down a word when un-

certain of its spelling. In fact, it has been shown that

the storage of these MOI’s is so strong, that exposure

to “incorrect versions of [. . . ] particular words can

interfere with subsequent spelling accuracy” [p. 492]

(Brown, 1988). This is why, in general, it is better to

confront learners mainly with correct spellings, hence

avoiding discrimination exercises of correctly and in-

correctly spelled words. Additionally, in the spelling

acquisition process it may be helpful to highlight dif-

ﬁcult letter patterns within words and that are contin-

ually misspelled, as is suggested by Arter (Arter and

Mason, 1994). This idea is realised in eSpindle, the

CAVL tool that was mentioned earlier. Feedback for

misspelled words locates the exact position of errors

by showing correct letters and replacing wrong letters

by lines.

2.3 Challenges for Blind Language

Learners

Blind and visually impaired people often have a well-

trained memory to make up for the lack of vision

(Couper, 1996). This often makes them superior lan-

guage learners as they can easily retain vocabulary.

However, especially in the L2 classroom, they may

also be “notoriously bad spellers” [p. 9] (Couper,

1996). Learning the orthography of new language

requires blind students to learn a new variation of

Braille script of that language. Letters in Braille of-

ten have differing representations in other languages,

i.e. accents in French and umlauts in German. Braille

also commonly uses shortened words to save space

and these abbreviations vary in different languages.

Blind students thus face quite a challenge when learn-

ing how to spell in a different language; mental ortho-

graphic images of words are only accessible in com-

bination with mastering a new Braille variation. How-

ever difﬁcult this may be, Braille is often the per-

ferred choice medium for blind students who study

vocabulary, as it allows them to read with their hands

CSEDU 2010 - 2nd International Conference on Computer Supported Education

502

just as sighted people use their eyes. Unfortunately,

there is “a great shortage of [printed] materials suit-

able for learners requiring non-visual methods” [p. 1]

(Boguslaw, 2000) and, what might be even more of a

handicap, if a teacher or a student wishes to use dif-

ferent learning material, it has to be made available

ﬁrst. One solution to this problem might be to use the

Braille line of a PC, a piece of hardware that trans-

lates the screen into embrossed writing, one line at a

time. However, the Braille line makes navigation and

skipping through information rather difﬁcult which is

why blind computer users often prefer to work with

text-to-speech technology. Not only is such software

universally available on most PC’s, it allows work-

ing with texts at the same speed as sighted users and

even much faster (Moos and Trouvain, 2007). While

the blessings of this development are undeniable for

the blind community, there are two major downsides

when working with screen readers. First, “simply re-

placing one interaction mode, such as the display of

text on a screen with a functionally equivalent mode

as in speaking thext aloud, is not necessarily equiv-

alent from the point of view of user experience” [p.

65] (Shinohara, 2009). So blind people miss out on

some of the information presented on screen. And

second, exposure to the written word is lacking al-

most entirely. Spelling acquisition must then become

even more of a conscious effort for blind language

learners and with available CAVL tools aimed primar-

ily at sighted users, the design of such software must

also be conscious if it is to include blind users. A

so-called ‘design for all’ approach might not be the

ideal solution, however, since “[p]roviding access to

people with certain types of disability can make the

product signiﬁcantly more difﬁcult to use by people

without disabilities, and often impossible to use by

people with a different type of disability” (Newell and

Gregor, 2000) as cited in (Shinohara, 2009).

2.4 Problem Formulation

Summarising the research ﬁndings from the previous

sections allows formulating the problem more con-

cretely. Language learners are well advised to study

vocabulary, that is meaning and spelling of words,

in a focused and decontextualised fashion. CAVL

tools may support learners in this process. Blind lan-

guage learners, who face great challenges when ac-

quiring orthography, cannot access the orthographic

feedback in existing CAVL applications to the same

extent as sighted learners. They have to rely either

on the Braille line, which is inconvenient since it is

slow and an extra piece of hardware, or on a screen

reader, which is suboptimal for information such as

orthographic feedback that is designed to be conveyed

visually. A nearby solution would then be to design

orthographic feedback to be conveyed via the audi-

tory channel. Katz has shown that “the many similar-

ities found in recognising orally spelled words and in

reading suggest that the abilities depend on the same

[lexical and] nonlexical procedures” [p. 215] (Katz,

1989). Katz explains that when recognising orally

spelled words, the letters are processed serially rather

than simultaneously as in reading. This indicates that

people piece together single letters to form a word in

their mind. Orthographic feedback in CAVL appli-

cations might then be given in form of oral spelling

when vision is impaired.

3 METHODOLOGY

3.1 Three Phase Approach

The goal of this work is to create a tool that is highly

usable for blind users, i.e. a tool that is useful, efﬁ-

cient, effective, satisfying, and accessible according

to Rubin’s deﬁnition of usability [p. 16] (Rubin and

Chisnell, 2008). This is why a user-centered-design

(UCD) approach has been adopted to construct a sys-

tem architecture in collaboration with users. “When

a product or service is truly usable, the user can do

what he or she wants to do the way he or she expects

to be able to do it, without hindrance, hesitation, or

questions” [p. 4] (Rubin and Chisnell, 2008). Us-

ability thus implies some sense of invisibility, when

something is usable it serves as an intuitive tool that

allows people to accomplish their tasks easily and ef-

fortlessly.

The UCD approach that has been chosen for this

study consists of three phases, where the last two rely

heavily on user involvement. The ﬁrst phase serves

merely to get a clear understanding of the problem

and gather knowledge from related studies. In the

second phase, the architecture is designed and im-

mediately tested with a user in a qualitative evalu-

ation. The architecture is then adapted accordingly

and tested again until a satisfactory design is found.

The last phase consists of implementing the ﬁnal ar-

chitecture and evaluating it under sophisticated test-

ing conditions. This model has been adopted from

Microsoft’s UCD approach (Harper et al., 2008) that

suggests a cycle of the following ﬁve phases: under-

standing, study, design, build, evaluate. However, the

aim of this work is merely to evaluate the concept of

auditory orthographic feedback which when applying

Microsoft’s design cycle, would mean completing the

IMPROVING SPELLING SKILLS FOR BLIND LANGUAGE LEARNERS - Orthographic Feedback in an Auditory

Vocabulary Trainer

503

cycle only once. To allow iterative testing neverthe-

less, the approach has been adopted as described.

3.2 Completed Phases

The three phases include understanding the problem,

designing a solution in collaboration with users, and

ﬁnally building and evaluating a demo system. All

three phases have been completed at the time of writ-

ing. First, the problem has been explored in the con-

text of related research as was described in section

2. Secondly, a preliminary solution has been tested

and gradually improved with seven target users, with

some users participating in the study more than once.

Thirdly, a demo version has been implemented in

JAVA and ﬁnally been evaluated with 15 pupils from

a special school for the blind.

In the following, section 4 describes the system

architecture. Section 5 presents the results from the

ﬁnal evaluation phase.

4 SYSTEM ARCHITECTURE

4.1 Conceptual Architecture

The goal of the design is to make orthographic feed-

back within a CAVL application fully available to

blind computer users who work with text-to-speech

technology. Drawing on the research ﬁndings dis-

cussed in section 2, the design should satisfy the fol-

lowing system requirements:

1. employ a ﬂash card style study method

2. give positive and encouraging feedback and limit

negative feedback

3. confront the user only with correct spellings and

point out the position of errors

4. function with a normal keyboard and TTS tech-

nology

5. achieve the same learning outcome as visual or-

thographic feedback would

In eSpindle, spelling mistakes are corrected as fol-

lows: when the user enters a misspelled word, for ex-

ample ‘a[k]qui[z]ition’, the word is deleted immedi-

ately from the screen and ‘a qui ition’ is displayed

instead. The user is then asked to enter the word

again. While this could be translated one-to-one us-

ing a screen reader, it is not very practical. When

‘a

qui ition’ is spelled auditorily without visual feed-

back, one has to remember the position of the errors,

ﬁgure out which letters were right and which ones

were wrong, and ﬁnally ﬁnd the correct letter—all in

the mind! So even though blind students tend to be

mind jugglers, this exercise is not intended to improve

memory skills but rather orthographic skills. So how

could orthographic feedback be given auditorily in a

way that serves its purpose better? This was the guid-

ing question in phase II (cp. section 3) of the design

construction. The main result was that people who

orally correct spelling mistakes for each other, slow

down where the error had occured and use a differ-

ent pitch. ‘a[k]qui[z]ition’ will most likely be cor-

rected by spelling the word in its appropriate form,

‘a[c]qui[s]ition’, and highlighting the positions of the

errors, in this case [c] and [s]. People also point di-

rectly to the mistake by saying ‘acquisition’ is spelled

with ‘c’, not a ‘k’. However, these types of correc-

tions vary greatly with the type of mistake and sim-

ply spelling out the entire word is a safe bet when it

comes to design consistency. The following reports

the architecture that is in line with the system require-

ments and was considered most usable by the seven

test subjects who took part in the design. Figure 4.1

illustrates the process.

The system initiates the interaction by giving the

German deﬁnition of a word (Gi). It then responds to

three types of user input, namely:

Correct: the system gives positive feedback and

moves on to the next word.

Not correct: the system gives short and unintrusive

negative feedback and says the word in German

(Gi) and English (Ei). The user can then attempt

to spell the word.

Partially correct: the system gives encouraging

feedback and spells the word by changing the

pitch where errors occurred (prosodically en-

hanced spelling). The user can then enter the word

anew.

In case the user needs more then one attempt, the

system adopts the response for ‘not correct’ as fol-

lows’:

Not correct: the system gives short and unintrusive

negative feedback, pronounces the word in En-

glish (Ei), and spells the entire word. The user

can then enter the word.

The system only moves on to the next word when

the correct answer has been given. The user can at

any point in the interaction access the German word

by typing ‘G’, the correct English translation by typ-

ing ‘E’, and the spelled version of the English word

by typing ‘S’. Words are classiﬁed as ‘learnt’ when

they are translated and spelled correctly after the ﬁrst

prompt, otherwise they are repeated in a loop.

CSEDU 2010 - 2nd International Conference on Computer Supported Education

504

Figure 1: Flowchart of the System Architecture.

4.2 Implementation

The goal of the demo version is to evaluate the ba-

sic concept of auditory orthographic feedback as it

has been descibed in the previous section. A demo

version of this architecture will be implemented in

JAVA. The vocabulary lists and the spelled words

will be read out by a by a text-to-speech (TTS) syn-

thesiser while the feedback is given by a recorded

voice. TTS technology is used because it eventu-

ally allows easy manipulation of words, such as slow-

ing down when repeating a word and changing pitch

when pointing to mistakes. It is the medium that blind

people commonly work with. This ensures also that

new words could eventually be added easily to the

database but has the downside of occasionally mis-

pronouncing words. The other feedback is given by

a human. It is assumed to sound more encouraging

when a person says “Well done!” with an enthusias-

tic voice as opposed to a robot that does not convey

emotions.

A simple spell checker that can ﬁnd omissions

(a quisition), insertions (acquis[s]ition) and substitu-

tions (a[k]quisition) will be used. It will classify a

word as correct if it has no mistakes, as partially cor-

rect if it has at most three mistakes, and as incorrect if

it has more than three mistakes. The number of mis-

takes is measured via Levenshtein distance.

5 EVALUATION

The evaluation of the ﬁnal architecture took place at a

special school for the blind with 15 participants, aged

12 to 16. The pupils were native speakers of Ger-

man, computer literate and had at least two years of

formal English education. The vocabulary test was

conducted following the between-group-design of tra-

ditional experiments, which includes a pretest and

posttest. Test participants were randomly assigned

one of two conditions: simple auditory feedback (S-

AF) and prosodically enhanced auditory feedback

(PE-AF). Every participant also completed two brief

questionnaires, one at the beginning of the session

about biographical information and study habits, and

one at the end of the session to evaluate the usability

of the system. Out of the 15 test participants, 5 also

took part in a delayed posttest to evaluate vocabulary

retention.

At the time of writing, the study was just com-

pleted and data has not been analysed yet. However,

the tendencies are very encouraging. There is a huge

learning gain in terms of spelling and word meaning

in both testing conditions when comparing the pretest

and the posttest at a glance. However, it must be

noted that this is also due to the effect of the short

term memory. Another interesting result is that most

of the pupils greatly enjoyed working with the sys-

tem and some even asked where they could buy it. 14

out 15 participants answered ‘yes’ to the question if

they would use such a program in their own time to

study vocabulary. In order to truly evaluate the effec-

tiveness of the system with regard to word retention,

a longitudinal study over the course of several weeks

would be appropriate as it would allow to integrate a

learning model of ‘timed repetitions’. Moreover, fu-

ture studies should include a control group in which

test subjects learn the same words using their tradi-

tional study methods.

IMPROVING SPELLING SKILLS FOR BLIND LANGUAGE LEARNERS - Orthographic Feedback in an Auditory

Vocabulary Trainer

505

6 CONCLUSIONS

It has been argued that orthographic feedback within

computer assisted vocabulary learning (CAVL) appli-

cations should be specially designed to increase us-

ability for blind language learners. Learning orthog-

raphy and in particular orthography of a foreign lan-

guage may be very challenging for visually impaired

people since it often implies learning a different type

of Braille script along with it. This is why well-

designed auditory feedback may be an alternative so-

lution to teach spelling within CAVL applications. A

system architecture was constructed in collaboration

with blind students and served as basis for the im-

plementation of a demo system. Finally, this demo

system has been evaluated with 15 pupils from a spe-

cial school for the blind, producing very encouraging

results.

Further research should address the content of the

vocabulary trainer as well evaluation methods. While

this study purely served to test an idea, that of prosod-

ically enhanced auditory feedback, it would be very

interesting to test this concept with more elaborated

vocabulary such as idioms or sentences. Also, one

could consider to test the effect of different types of

encouraging feedback that is given when a word is en-

tered correcty. In order to evaluate the true effect of

this learning method, a study would ideally last sev-

eral sessions and include a control group that learns

vocabulary using traditional revision techniques.

REFERENCES

Arter, C. and Mason, H. (1994). Spelling for the visually

impaired child. British Journal of Visual Impairment,

12(18).

Boguslaw, M. (2000). Learning from experience. mobility

and daily living skills in an english language class-

room. Technical report, Dept. of ELT Typhlomethod-

ology, Catholic University of Lublin.

Brown, A. S. (1988). Encountering misspellings and

spelling performance: Why wrong isn’t right. Jour-

nal of Educational Psychology, 4(488-494).

Cooley, R. E. (2001.). Vocabulary acquisition software:

user preferences and tutorial guidance. In AIED

2001 Workshop Papers: Computer Assisted Language

Learning.

Couper, H. (1996). Teaching modern languages to visually

impaired children. Language Learning Journal, 13.

eSpindle (2009). espindle homepage. http://www. espin-

dle.org/ Last checked on 21 December 2009.

Goodfellow, R. (1995). A review of the types of call pro-

grams for vocabulary instruction. Computer Assisted

Language Learning, 8(2-3):205–226.

Groot, P. (2000). Computer assisted second language vo-

cabulary acquisition. Language Learning and Tech-

nology, 4(1):60–81.

Harper, R., Rodden, T., Rogers, Y., and Sellen, A., ed-

itors (2008). Being Human: Human-Computer In-

teraction in the Year 2020. Microsoft Research Ltd.

http://research.microsoft.com/hci2020/.

Katz, R. B. (1989). Recognizing orally spelled words:

An analysis of procedures shared with reading and

spelling. BRAIN AND LANGUAGE, 37:201–219.

Ma, Q. and Kelly, P. (2006). Computer assisted vocabulary

learning: design and evaluation. Computer Assisted

Language Learning, 19(1):15–45.

Moos, A. and Trouvain, J. (2007). Comprehension of

ultra-fast speech—blind vs. “normally hearing” per-

sons. 16th International Congress of Phonetic Sci-

ences, Saarbr

ucken,, pages pp. 677–680.

Newell, A. and Gregor, P. (2000). User-sensitive inclusive

design—In search of a new paradigm. ACM Press,

Arlington, VA.

Olmanson, J. (2007). Review of espindle vocabulary and

spelling program online. Language Learning and

Technology, 11(3):18–28.

Phase6 (2009). Phase 6 homepage. http://www.phase-

6.de/opencms/Homepage/ Last checked on 21 De-

cember 2009.

Rubin and Chisnell (2008). Handbook of Usability Testing,

Second Edition: How to Plan, Design, and Conduct

Effective Tests. Wiley Publishing, Indianapolis, Indi-

ana.

Shinohara, K. (2009). A blind person’s interactions with

technology. Communications of the ACM, 52(8):58–

66.

Torbe, M. (1977). Teaching Spelling. Ward Lock Education

Ltd., London.

Warda, R. (2009). Research based tutoring of

english spelling: White paper (research).

http://www.espindle.org/whitepaper.pdf Last checked

on 16 September 2009.

Warschauer, M. (1996). Computer Assisted Language

Learning: an Introduction. Logos International. Fo-

tos S. (ed.), Tokyo.

CSEDU 2010 - 2nd International Conference on Computer Supported Education

506