Learning false friends across contexts

Abstract

False friends are words in two languages that look or sound similar but differ significantly in meaning in some or all contexts. False friends are confusing for language students and could result in frustration and communication problems. This paper proposes a method to diagnose and prevent false friends mistakes based on students' past learned words, current location and time. The proposed method uses records from the SCROLL system (System for Capturing and Reminding Of Learning Log) to analyze the previous activity of students. We assume that the past activity of a student can be used to predict the meaning intended by the student when looking up a polysemous word. The identification of the intended meaning in the student's current context is then used to provide the student with the appropriate translation, warnings and quizzes, possibly improving the learning process and avoiding false friends future mistakes.

Keywords

• Learning Analytics
• Ubiquitous learning
• False Friends
• Computer Supported Language Learning

INTRODUCTION

When learning a second language, students can take advantage of the vocabulary of their first language using cognates (Nation, 2003). Cognates are words that sound or look similar in the two languages, have similar meanings, and help students expand their vocabulary by playing the role of `true friends'.

apartment). The previous situations are faced by language learners and are caused by a tricky category of words: false friends.

False friends are words in two languages that look and sound similar, but differ significantly in meaning in some or all contexts. The degree of complexity of learning false friends depends on whether they are total false friends or partial false friends. Total false friends have completely different meanings in both languages (e.g.: Eng.: Attend (to be present); French: Attendre (wait)), whereas partial false friends are polysemous words, one of whose meanings is a false friend while others are true cognates (e.g.: Eng.: Demand (i.e., request made as of right); French: Demander (i.e., to ask; to be looking for; to demand). Figure 1 shows the types of `friends' that a student will encounter when learning the target language.

Depending on the context, partial false friends play the role of a true or a false friend. This paper

proposes a method to diagnose and prevent false friends mistakes based on students' past learned vocabulary, current location and time. First, we analyze the factors that lead to false friends confusions while learning the target language. We also analyze the types of assimilation problems that could then arise, depending on the type of false friend. In order to prevent false friends confusions, we use the student's past learning logs (previous looked up words, locations, time) to predict the meaning intended by the student when looking up a polysemous words. The identification of the intended meaning is then used to provide the student with the appropriate translation and warnings, possibly improving the learning process and avoiding future mistakes. In order to fortify the learning the student knowledge will be tested. The language learner will get quizzes about the meaning of learned polysemous words in different contexts (Location, time). 2

DIFFICULTIES AND PROBLEMS OF FALSE FRIENDS LEARNING

From the pedagogical perspective, intrinsic and extrinsic factors determine the degree of false friends' difficulty for language (Beltran, 2006). The intrinsic factors contributing to a higher level of difficulty of false friends learning are:

IF1: The confusing nature of false friends, and particularly the partial false friends. Some false friends have always a deceptive meaning, whereas some others have deceptive meanings in certain contexts only. This creates uncertainty for students as they could fail to recognize in which contexts the word is a false friend, and in which contexts it is not.

IF2: Semantic fields may overlap. Words can have different meanings in both languages, but belong to the same semantic field (e.g.: Japanese: フィルム. firumu means camera film roll). IF3: Because of the large number of true friends, students have a tendency to overgeneralize the words that they come across.

The extrinsic factors contributing to the complexity of false friends learning are:

EF1: Language learners are usually encouraged to take advantage of true cognates without being warned of the existence of false friends. This could lead to frustration for the language learner when they notice the actual complexity of the cognates.

EF2: Oversimplification of dictionaries by lexicographers where translations sometimes lack of nuances and contexts.

Depending on the type of false friend, the previous factors influence differently the difficulty in

learning them. In order to improve false friends teaching, it is important to understand which type of factors influence false friends learning. Table 1 shows which factors affect which type of false friends learning as well as the overall learning difficulty. X is displayed when a type of false friends learning is not affected by a factor. O is displayed when a type of false friend is affected by a factor.

IF1 IF2 IF3 EF1

EF2

Difficulty X

Medium O

High O

High

O Low False friends are rarely incorporated into language classes despite the difficulties faced by language students when dealing with them. When they are pointed out by the teachers, the words’ nuances are often over-simplified and downgrade the accuracy and assimilation of the meaning of the word (Hayward, 1984) . In the case of partial false friends, this lack of accuracy can lead to two different situations: • • a loss of some meanings of the word in the target language. an addition of some meanings to the word in the target language by projecting the meaning of the word in the native language to the word in the target language.

In order to avoid this kind of mistakes, false friends need to be pointed out by the teachers and the correct translations should be given (August, 2005). Moreover, the meanings should not be

presented as one bloc to the student but should be put in a context and restricted to a particular area (Hayward, 1984) .

However, in a self learning environment, teachers have less or no control over the learned content.

With the rise of smart phones, the use of mobile devices in language learning is a growing trend (Godwin, 2016). Student have more control over their learning pace (Benson, 2005) and the vocabulary they want to learn, but don't have opportunities to communicate which meaning of the word they are looking for.

This paper proposes to take advantage of student past learned words to understand the particular meaning queried by the learner and provide them with the correct translation in their intended context. Moreover, we aim at minimizing the effect of the intrinsic and extrinsic factors affecting false friends learning by showing the learner the different meanings and nuances of the words. The assimilation of the student will be then tested in the context of usage of the word in order to fortify their knowledge. 3 3.1

METHOD

SCROLL system

During this study we use records from the SCROLL System (System for Capturing and Reminding Of

Learning Log). Scroll is a digital record of what language students have learned in daily life. It allows

the learners to log the new words or sentences they learned along with photos, audios, videos and location (Ogata, 2011) . SCROLL captures what learners are learning as well as its contextual data.

The users are then reminded of what they learned in the right place and the right time. Moreover,

students receive personalized quizzes to fortify the learning. Figure 3 is a screenshot from the

SCROLL system that shows a log inserted by a student for the word Karaoke. The student appended

a picture and a location when creating the log. A Japanese translation of the word Karaoke is automatically provided to the student, and the time is automatically registered.

The logs we will be using include meta-data such as:

Knowledge: words that students have learned in the past User: author identification Place: place where the learning happened (cinema, restaurant) Time: time when the learning took place

Currently SCROLL has 1648 users and contains 24355 logs. The system is used mainly by students learning Japanese.

3.2

Contextual false friends learning When using the SCROLL system, Japanese language learners insert logs containing a word in English and learn its Japanese translation. However, if the word is a false friend, students get a translation that does not usually reflect the context, the different meanings and the nuances of the word. In order to provide learners with the right translation in the right context, we have to understand their intended meaning. We propose to analyze the past activity of students to extract the meaning they are looking for. We assume that the past activity of a student gives a context that can be used to extract the meaning intended by the student when looking up a polysemous word. The following meta-data are used to predict the meaning the student is looking for: Past knowledge (words): to measure the similarity between the current word the student is looking up and past words the student looked up in the system.

Time: to measure the similarity between the current word the student is looking up and the set of words the student looked up at the same period of time.

Location: to measure the similarity between the word the student is looking up and the set of words the student looked up at the same location.

We propose to measure the similarity using Jiang similarity measure. Jiang similarity measure based

information content of each concept in WordNet. It assumes that each concept includes information in WordNet and the more common information two concepts share, the more similar the two concepts are (Meng, 2013) . Previous studies where semantic distance measures were compared experimentally found that Jiang's measure gave the best results overall (Budanitsky, 2001) . Figure 4 shows an example. The student looked up the word driver. However, the word driver has different meanings and different translations in Japanese for each of those meaning:

Person who drives a vehicle: ドライバ (doraiba)

Train driver: 運転⼠ (untenshi) Mass driver : マスドライバー (masu doraiba) Computing : 仮想デバイスドライバ (kasoo debaisu doraiba) Golf club: ドライバー (doraiba) Screwdriver (British English):ドライバー (doraiba) In order to understand which meaning the student is looking for, we look at the previous knowledge, the words looked up at the same period of time than the word driver and the words looked up at the same location of the word driver. In this case, the word looked up just before the word driver is stepladder. The semantic distance between the words srewdriver and stepladder is smaller than the semantic distance between the word stepladder and each of the words: car driver, train driver, computer driver, mass driver and golf club. We can conclude that the meaning the student is looking for is screwdriver, and that the translation that should be given is ドライバー (doraiba). 3.3

Preventive false friends learning

As stated previously, students face intrinsic and extrinsic factors that make false friend learning

more difficult. We propose to tackle the factors by providing preventive false friends learning. Upon the encounter of a new false friend, the student will receive a warning that shows a comprehensive list of translations in different contexts. The displayed warning is different for each type of false friends as shown in table 2.

False Friend Type

F2: False Friends

F3: Partial False Friends

N∩T

Warning stating that the word is a false friend.

Warning that:

•

states that the word is a partial false friend. • lists the different translations for the different meanings of the word. • lists the cognates of the word in the target language that have meanings that are non-existent in the native language. F4: Partial False Friends • states that the word is a partial false friend.

T⊂N

• item lists the different translations for the different meanings of the word. F5: Partial False Friends • states that the word is a partial false friend.

N⊂T • lists the cognates of the word in the target language that have meanings that are non-existent in the native language.

The expected consequences of displaying warnings are:

Consequence1: Awareness that the words are used differently in different contexts: The warning states the false friend type.

Consequence2: Knowledge that avoids over-generalization (addition of meanings) : The warning lists the different translations in different contexts.

Consequence3: Knowledge that avoids over-simplification (loss of meanings): The warning lists the different translations in different contexts.

In the previous example, the student looked up the word driver. After predicting the intended

meaning of the student, the student will be provided with the word ドライバー (doraiba) as a translation. However, driver and ドライバー are partial false friends of type 4 (F4). In this case, the student might use the word ドライバー (doraiba) as they would have used the word driver in

English to express Train driver, mass driver or driver in computing. To avoid this mistake, we propose

to display a warning to the student showing the different translations of the word driver, depending on the meaning (as shown in Figure 5). The warning gives the student awareness about the complexity of the word, and consciousness that the translation provided by the system can be used in particular contexts only. In order to determine the effect of contextual translations and warnings on false friends learning, students will be given quizzes to test their acquired knowledge. SCROLL system offers the opportunity to give quizzes to students depending on their location. When the students will be present at a location related to one of the meanings of the word a quiz will appear asking the student about the translation of the word in this particular context as shown in figure 6. The quizzes will be given to the group of students that received the contextual translation as well as the warnings. Another group of students that didn't get the contextual translation and the warnings will play the role of a control group and be subjected to quizzes as well. The results will be compared to identify the effect of the previous method on false friends learning.

DISCUSSION

We proposed a method to prevent false friends mistakes. The method is applied in the context of

mobile learning. The main features of mobile learning are accessibility, immediacy, interactivity and situating of instructional activities (Ogata, 2004) , benefit students during the learning process.

However, mobile learning does not allow language students to share their intended meaning while

looking up a polysemous word. The first part of our work consists of predicting the meaning intended by the student when they are looking up a polysemous word. In order to do so, the proposed method uses records from the SCROLL system (System for Capturing and Reminding Of

Learning Log) to analyze the previous activity of students. We assumed that the students' past

learned words, current location and time can be used to predict the meaning intended by the student when looking up a polysemous word. The identification of the intended meaning is then used to provide the student with the appropriate translation, based on the intended meaning. The second part proposes to display warnings and quizzes to the students. The warnings explain the meaning of the word and provide the student with different translations in the different contexts.

The quizzes aim at fortifying the knowledge of the students, possibly improving the learning process.

This method puts into application the theoretical pedagogical approach of false friends learning.

Future work will evaluate the accuracy of the prediction of the students' intended meaning. We will

also evaluate the effects of the warnings by comparing the learning performance of the students Creative Commons License, Attribution - NonCommercial-NoDerivs 3.0 Unported (CC BY-NC-ND 3.0) before and after receiving them. The impact of the quizzes will be evaluated by comparing the false friends recall rate of the student before and after the display of the quizzes. 5

CONCLUSION

This paper proposes a method to diagnose and prevent false friends mistakes based on students’

past learning activity. The proposed method uses records from the SCROLL system (System for

Capturing and Reminding Of Learning Log) to analyze the previous activity of students. We assumed

that the students' past learned words, current location and time can be used to predict the meaning intended by the student when looking up a polysemous word. The identification of the intended meaning is then used to provide the student with the appropriate translation, warnings and quizzes, possibly improving the learning process and avoiding future mistakes.

Acknowledgments

References

This research was also partially supported by Grant-in-Aid for Scientific Research (S) from Japan Society for the Promotion of Science ( 24220002 , 2012 - 2016 ) and from Monbukagakusho ( 16H06304 , 2016 - 2021 ).

August , D. , Carlo , M. , Dressler , C. , and Snow , C. ( 2005 ). The critical role of vocabulary development for english language learners . Learning Disabilities Research & Practice , 20 ( 1 ): 50 - 57 .

Beltrán , R. C. ( 2006 ). Towards a typological classification of false friends (spanish-english) . Resla , 19 : 29 - 39 .

Benson , P. ( 2005 ). Autonomy and information technology in the educational discourse of the information . Information technology and innovation in language education , 1 : 173 .

Budanitsky , A. and Hirst , G. ( 2001 ). Semantic distance in wordnet: An experimental, applicationoriented evaluation of five measures . In Workshop on WordNet and other lexical resources , volume 2 , pages 2 - 2 .

Godwin-Jones , R. ( 2016 ). Looking back and ahead: 20 years of technologies for language learning .

Hayward , T. and Moulin , A. ( 1984 ). False friends invigorated . In Lexeter'83 Proceedings , pages 190 - 198 .

Kay , G. ( 1995 ). English loanwords in japanese . World Englishes , 14 ( 1 ): 67 - 76 .

Meng , L. , Huang , R. , and Gu , J. ( 2013 ). A review of semantic similarity measures in wordnet . International Journal of Hybrid Information Technology , 6 ( 1 ): 1 - 12 .

Nation , P. ( 2003 ). The role of the first language in foreign language learning . Asian EFL Journal , 5 ( 2 ): 1 - 8 .

Ogata , H. , Li , M. , Hou , B. , Uosaki , N. , El-Bishouty , M. M. , and Yano , Y. ( 2011 ). Scroll: Supporting to share and reuse ubiquitous learning log in the context of language learning . Research & Practice in Technology Enhanced Learning , 6 ( 2 ).

Ogata , H. and Yano , Y. ( 2004 ). Knowledge awareness for a computerassisted language learning using handhelds . International Journal of Continuing Engineering Education and Life Long Learning , 14 ( 4-5 ): 435 - 449 .